id sid tid token lemma pos bracis-28410 1 1 single single ADJ bracis-28410 1 2 image image NOUN bracis-28410 1 3 super super NOUN bracis-28410 1 4 - - NOUN bracis-28410 1 5 resolution resolution NOUN bracis-28410 1 6 based base VERB bracis-28410 1 7 on on ADP bracis-28410 1 8     SPACE bracis-28410 1 9 capsule capsule NOUN bracis-28410 1 10 neural neural ADJ bracis-28410 1 11 networks network NOUN bracis-28410 1 12 | | ADV bracis-28410 1 13 springer springer NOUN bracis-28410 1 14 nature nature PROPN bracis-28410 1 15 link link PROPN bracis-28410 1 16 ( ( PUNCT bracis-28410 1 17 formerly formerly ADV bracis-28410 1 18 springerlink springerlink NOUN bracis-28410 1 19 ) ) PUNCT bracis-28410 1 20 skip skip VERB bracis-28410 1 21 to to ADP bracis-28410 1 22 main main ADJ bracis-28410 1 23 content content NOUN bracis-28410 1 24 advertisement advertisement NOUN bracis-28410 1 25 log log NOUN bracis-28410 1 26 in in ADP bracis-28410 1 27 menu menu NOUN bracis-28410 1 28 find find VERB bracis-28410 1 29 a a DET bracis-28410 1 30 journal journal NOUN bracis-28410 1 31 publish publish VERB bracis-28410 1 32 with with ADP bracis-28410 1 33 us we PRON bracis-28410 1 34 track track VERB bracis-28410 1 35 your your PRON bracis-28410 1 36 research research NOUN bracis-28410 1 37 search search NOUN bracis-28410 1 38 cart cart NOUN bracis-28410 2 1 home home NOUN bracis-28410 2 2 intelligent intelligent ADJ bracis-28410 2 3 systems system NOUN bracis-28410 2 4 conference conference NOUN bracis-28410 2 5 paper paper NOUN bracis-28410 2 6 single single ADJ bracis-28410 2 7 image image NOUN bracis-28410 2 8 super super NOUN bracis-28410 2 9 - - NOUN bracis-28410 2 10 resolution resolution NOUN bracis-28410 2 11 based base VERB bracis-28410 2 12 on on ADP bracis-28410 2 13     SPACE bracis-28410 2 14 capsule capsule PROPN bracis-28410 2 15 neural neural ADJ bracis-28410 2 16 networks network NOUN bracis-28410 2 17 conference conference NOUN bracis-28410 2 18 paper paper NOUN bracis-28410 2 19 first first ADV bracis-28410 2 20 online online ADV bracis-28410 2 21 : : PUNCT bracis-28410 2 22 12 12 NUM bracis-28410 2 23 october october NOUN bracis-28410 2 24 2023 2023 NUM bracis-28410 2 25 pp pp ADP bracis-28410 2 26 113–128 113–128 NUM bracis-28410 2 27 cite cite VERB bracis-28410 2 28 this this DET bracis-28410 2 29 conference conference NOUN bracis-28410 2 30 paper paper NOUN bracis-28410 2 31 access access NOUN bracis-28410 2 32 provided provide VERB bracis-28410 2 33 by by ADP bracis-28410 2 34 university university PROPN bracis-28410 2 35 of of ADP bracis-28410 2 36 notre notre PROPN bracis-28410 2 37 dame dame PROPN bracis-28410 2 38 hesburgh hesburgh PROPN bracis-28410 2 39 library library PROPN bracis-28410 2 40 download download PROPN bracis-28410 2 41 book book NOUN bracis-28410 2 42 pdf pdf PROPN bracis-28410 2 43 download download NOUN bracis-28410 2 44 book book NOUN bracis-28410 2 45 epub epub PROPN bracis-28410 2 46 intelligent intelligent ADJ bracis-28410 2 47 systems system NOUN bracis-28410 2 48 ( ( PUNCT bracis-28410 2 49 bracis bracis NOUN bracis-28410 2 50 2023 2023 NUM bracis-28410 2 51 ) ) PUNCT bracis-28410 2 52 single single ADJ bracis-28410 2 53 image image NOUN bracis-28410 2 54 super super NOUN bracis-28410 2 55 - - NOUN bracis-28410 2 56 resolution resolution NOUN bracis-28410 2 57 based base VERB bracis-28410 2 58 on on ADP bracis-28410 2 59     SPACE bracis-28410 2 60 capsule capsule NOUN bracis-28410 2 61 neural neural ADJ bracis-28410 2 62 networks network NOUN bracis-28410 2 63 download download VERB bracis-28410 2 64 book book NOUN bracis-28410 2 65 pdf pdf PROPN bracis-28410 2 66 download download NOUN bracis-28410 2 67 book book PROPN bracis-28410 2 68 epub epub PROPN bracis-28410 2 69 george george PROPN bracis-28410 2 70 corrêa corrêa PROPN bracis-28410 2 71 de de PROPN bracis-28410 2 72 araújo9 araújo9 PROPN bracis-28410 2 73 , , PUNCT bracis-28410 2 74 artur artur PROPN bracis-28410 2 75 jordão10 jordão10 PROPN bracis-28410 2 76 & & CCONJ bracis-28410 2 77 helio helio PROPN bracis-28410 2 78 pedrini9 pedrini9 PROPN bracis-28410 2 79     SPACE bracis-28410 2 80 part part NOUN bracis-28410 2 81 of of ADP bracis-28410 2 82 the the DET bracis-28410 2 83 book book NOUN bracis-28410 2 84 series series NOUN bracis-28410 2 85 : : PUNCT bracis-28410 2 86 lecture lecture NOUN bracis-28410 2 87 notes note NOUN bracis-28410 2 88 in in ADP bracis-28410 2 89 computer computer NOUN bracis-28410 2 90 science science NOUN bracis-28410 2 91 ( ( PUNCT bracis-28410 2 92 ( ( PUNCT bracis-28410 2 93 lnai lnai ADJ bracis-28410 2 94 , , PUNCT bracis-28410 2 95 volume volume NOUN bracis-28410 2 96 14197 14197 NUM bracis-28410 2 97 ) ) PUNCT bracis-28410 2 98 ) ) PUNCT bracis-28410 2 99 included include VERB bracis-28410 2 100 in in ADP bracis-28410 2 101 the the DET bracis-28410 2 102 following follow VERB bracis-28410 2 103 conference conference NOUN bracis-28410 2 104 series series NOUN bracis-28410 2 105 : : PUNCT bracis-28410 2 106 brazilian brazilian ADJ bracis-28410 2 107 conference conference NOUN bracis-28410 2 108 on on ADP bracis-28410 2 109 intelligent intelligent ADJ bracis-28410 2 110 systems system NOUN bracis-28410 2 111 529 529 NUM bracis-28410 2 112 accesses access NOUN bracis-28410 2 113 1 1 NUM bracis-28410 2 114 citation citation NOUN bracis-28410 2 115 abstract abstract ADJ bracis-28410 2 116 single single ADJ bracis-28410 2 117 image image NOUN bracis-28410 2 118 super super NOUN bracis-28410 2 119 - - NOUN bracis-28410 2 120 resolution resolution ADJ bracis-28410 2 121 ( ( PUNCT bracis-28410 2 122 sisr sisr NOUN bracis-28410 2 123 ) ) PUNCT bracis-28410 2 124 consists consist VERB bracis-28410 2 125 of of ADP bracis-28410 2 126 obtaining obtain VERB bracis-28410 2 127 one one NUM bracis-28410 2 128 high high ADJ bracis-28410 2 129 - - PUNCT bracis-28410 2 130 resolution resolution NOUN bracis-28410 2 131 version version NOUN bracis-28410 2 132 of of ADP bracis-28410 2 133 a a DET bracis-28410 2 134 low low ADJ bracis-28410 2 135 - - PUNCT bracis-28410 2 136 resolution resolution NOUN bracis-28410 2 137 image image NOUN bracis-28410 2 138 by by ADP bracis-28410 2 139 increasing increase VERB bracis-28410 2 140 the the DET bracis-28410 2 141 number number NOUN bracis-28410 2 142 of of ADP bracis-28410 2 143 pixels pixel NOUN bracis-28410 2 144 per per ADP bracis-28410 2 145 unit unit NOUN bracis-28410 2 146 area area NOUN bracis-28410 2 147 . . PUNCT bracis-28410 3 1 this this DET bracis-28410 3 2 method method NOUN bracis-28410 3 3 has have AUX bracis-28410 3 4 been be AUX bracis-28410 3 5 actively actively ADV bracis-28410 3 6 investigated investigate VERB bracis-28410 3 7 by by ADP bracis-28410 3 8 the the DET bracis-28410 3 9 research research NOUN bracis-28410 3 10 community community NOUN bracis-28410 3 11 , , PUNCT bracis-28410 3 12 due due ADP bracis-28410 3 13 to to ADP bracis-28410 3 14 the the DET bracis-28410 3 15 wide wide ADJ bracis-28410 3 16 variety variety NOUN bracis-28410 3 17 of of ADP bracis-28410 3 18 problems problem NOUN bracis-28410 3 19 ranging range VERB bracis-28410 3 20 from from ADP bracis-28410 3 21 real real ADJ bracis-28410 3 22 - - PUNCT bracis-28410 3 23 world world NOUN bracis-28410 3 24 surveillance surveillance NOUN bracis-28410 3 25 to to ADP bracis-28410 3 26 aerial aerial ADJ bracis-28410 3 27 and and CCONJ bracis-28410 3 28 satellite satellite NOUN bracis-28410 3 29 imaging imaging NOUN bracis-28410 3 30 . . PUNCT bracis-28410 4 1 most most ADJ bracis-28410 4 2 of of ADP bracis-28410 4 3 the the DET bracis-28410 4 4 improvements improvement NOUN bracis-28410 4 5 in in ADP bracis-28410 4 6 sisr sisr NOUN bracis-28410 4 7 come come VERB bracis-28410 4 8 from from ADP bracis-28410 4 9 convolutional convolutional ADJ bracis-28410 4 10 networks network NOUN bracis-28410 4 11 , , PUNCT bracis-28410 4 12 in in ADP bracis-28410 4 13 which which PRON bracis-28410 4 14 approaches approach NOUN bracis-28410 4 15 often often ADV bracis-28410 4 16 focus focus VERB bracis-28410 4 17 on on ADP bracis-28410 4 18 the the DET bracis-28410 4 19 deeper deep ADJ bracis-28410 4 20 and and CCONJ bracis-28410 4 21 wider wide ADJ bracis-28410 4 22 architectural architectural ADJ bracis-28410 4 23 paradigm paradigm NOUN bracis-28410 4 24 . . PUNCT bracis-28410 5 1 in in ADP bracis-28410 5 2 this this DET bracis-28410 5 3 work work NOUN bracis-28410 5 4 , , PUNCT bracis-28410 5 5 we we PRON bracis-28410 5 6 decided decide VERB bracis-28410 5 7 to to PART bracis-28410 5 8 step step VERB bracis-28410 5 9 up up ADP bracis-28410 5 10 from from ADP bracis-28410 5 11 the the DET bracis-28410 5 12 traditional traditional ADJ bracis-28410 5 13 convolutions convolution NOUN bracis-28410 5 14 and and CCONJ bracis-28410 5 15 adopt adopt VERB bracis-28410 5 16 the the DET bracis-28410 5 17 concept concept NOUN bracis-28410 5 18 of of ADP bracis-28410 5 19 capsules capsule NOUN bracis-28410 5 20 . . PUNCT bracis-28410 6 1 since since SCONJ bracis-28410 6 2 their their PRON bracis-28410 6 3 overwhelming overwhelming ADJ bracis-28410 6 4 results result NOUN bracis-28410 6 5 in in ADP bracis-28410 6 6 image image NOUN bracis-28410 6 7 classification classification NOUN bracis-28410 6 8 and and CCONJ bracis-28410 6 9 segmentation segmentation NOUN bracis-28410 6 10 problems problem NOUN bracis-28410 6 11 , , PUNCT bracis-28410 6 12 we we PRON bracis-28410 6 13 question question VERB bracis-28410 6 14 how how SCONJ bracis-28410 6 15 suitable suitable ADJ bracis-28410 6 16 they they PRON bracis-28410 6 17 are be AUX bracis-28410 6 18 for for ADP bracis-28410 6 19 sisr sisr NOUN bracis-28410 6 20 . . PUNCT bracis-28410 7 1 we we PRON bracis-28410 7 2 also also ADV bracis-28410 7 3 verify verify VERB bracis-28410 7 4 that that SCONJ bracis-28410 7 5 different different ADJ bracis-28410 7 6 solutions solution NOUN bracis-28410 7 7 share share VERB bracis-28410 7 8 similar similar ADJ bracis-28410 7 9 configurations configuration NOUN bracis-28410 7 10 , , PUNCT bracis-28410 7 11 and and CCONJ bracis-28410 7 12 argue argue VERB bracis-28410 7 13 that that SCONJ bracis-28410 7 14 this this DET bracis-28410 7 15 trend trend NOUN bracis-28410 7 16 leads lead VERB bracis-28410 7 17 to to ADP bracis-28410 7 18 fewer few ADJ bracis-28410 7 19 explorations exploration NOUN bracis-28410 7 20 of of ADP bracis-28410 7 21 network network NOUN bracis-28410 7 22 designs design NOUN bracis-28410 7 23 . . PUNCT bracis-28410 8 1 throughout throughout ADP bracis-28410 8 2 our our PRON bracis-28410 8 3 experiments experiment NOUN bracis-28410 8 4 , , PUNCT bracis-28410 8 5 we we PRON bracis-28410 8 6 check check VERB bracis-28410 8 7 various various ADJ bracis-28410 8 8 strategies strategy NOUN bracis-28410 8 9 to to PART bracis-28410 8 10 improve improve VERB bracis-28410 8 11 results result NOUN bracis-28410 8 12 , , PUNCT bracis-28410 8 13 ranging range VERB bracis-28410 8 14 from from ADP bracis-28410 8 15 new new ADJ bracis-28410 8 16 and and CCONJ bracis-28410 8 17 different different ADJ bracis-28410 8 18 loss loss NOUN bracis-28410 8 19 functions function NOUN bracis-28410 8 20 to to AUX bracis-28410 8 21 changes change NOUN bracis-28410 8 22 in in ADP bracis-28410 8 23 the the DET bracis-28410 8 24 capsule capsule NOUN bracis-28410 8 25 layers layer NOUN bracis-28410 8 26 . . PUNCT bracis-28410 9 1 our our PRON bracis-28410 9 2 network network NOUN bracis-28410 9 3 achieved achieve VERB bracis-28410 9 4 positive positive ADJ bracis-28410 9 5 and and CCONJ bracis-28410 9 6 promising promising ADJ bracis-28410 9 7 results result NOUN bracis-28410 9 8 with with ADP bracis-28410 9 9 fewer few ADJ bracis-28410 9 10 convolutional convolutional ADJ bracis-28410 9 11 - - PUNCT bracis-28410 9 12 based base VERB bracis-28410 9 13 layers layer NOUN bracis-28410 9 14 , , PUNCT bracis-28410 9 15 showing show VERB bracis-28410 9 16 that that SCONJ bracis-28410 9 17 capsules capsule NOUN bracis-28410 9 18 might might AUX bracis-28410 9 19 be be AUX bracis-28410 9 20 a a DET bracis-28410 9 21 concept concept NOUN bracis-28410 9 22 worth worth ADJ bracis-28410 9 23 applying apply VERB bracis-28410 9 24 to to ADP bracis-28410 9 25 the the DET bracis-28410 9 26 image image NOUN bracis-28410 9 27 super super ADJ bracis-28410 9 28 - - ADJ bracis-28410 9 29 resolution resolution ADJ bracis-28410 9 30 problem problem NOUN bracis-28410 9 31 . . PUNCT bracis-28410 10 1 in in ADP bracis-28410 10 2 particular particular ADJ bracis-28410 10 3 , , PUNCT bracis-28410 10 4 we we PRON bracis-28410 10 5 observe observe VERB bracis-28410 10 6 that that SCONJ bracis-28410 10 7 the the DET bracis-28410 10 8 proposed propose VERB bracis-28410 10 9 method method NOUN bracis-28410 10 10 recreates recreate VERB bracis-28410 10 11 the the DET bracis-28410 10 12 connection connection NOUN bracis-28410 10 13 between between ADP bracis-28410 10 14 the the DET bracis-28410 10 15 different different ADJ bracis-28410 10 16 characters character NOUN bracis-28410 10 17 more more ADV bracis-28410 10 18 precisely precisely ADV bracis-28410 10 19 , , PUNCT bracis-28410 10 20 thus thus ADV bracis-28410 10 21 demonstrating demonstrate VERB bracis-28410 10 22 the the DET bracis-28410 10 23 potential potential NOUN bracis-28410 10 24 of of ADP bracis-28410 10 25 capsules capsule NOUN bracis-28410 10 26 in in ADP bracis-28410 10 27 super super ADJ bracis-28410 10 28 - - ADJ bracis-28410 10 29 resolution resolution ADJ bracis-28410 10 30 problems problem NOUN bracis-28410 10 31 . . PUNCT bracis-28410 11 1 artur artur PROPN bracis-28410 11 2 jordáo jordáo PROPN bracis-28410 11 3 : : PUNCT bracis-28410 11 4 this this DET bracis-28410 11 5 work work NOUN bracis-28410 11 6 was be AUX bracis-28410 11 7 done do VERB bracis-28410 11 8 when when SCONJ bracis-28410 11 9 artur artur PROPN bracis-28410 11 10 jordao jordao PROPN bracis-28410 11 11 was be AUX bracis-28410 11 12 a a DET bracis-28410 11 13 post post ADJ bracis-28410 11 14 - - ADJ bracis-28410 11 15 doctoral doctoral ADJ bracis-28410 11 16 researcher researcher NOUN bracis-28410 11 17 at at ADP bracis-28410 11 18 the the DET bracis-28410 11 19 university university PROPN bracis-28410 11 20 of of ADP bracis-28410 11 21 campinas campinas PROPN bracis-28410 11 22 . . PUNCT bracis-28410 12 1 access access NOUN bracis-28410 12 2 provided provide VERB bracis-28410 12 3 by by ADP bracis-28410 12 4 university university PROPN bracis-28410 12 5 of of ADP bracis-28410 12 6 notre notre PROPN bracis-28410 12 7 dame dame PROPN bracis-28410 12 8 hesburgh hesburgh PROPN bracis-28410 12 9 library library PROPN bracis-28410 12 10 . . PUNCT bracis-28410 13 1 download download PROPN bracis-28410 13 2 conference conference NOUN bracis-28410 13 3 paper paper NOUN bracis-28410 13 4 pdf pdf NOUN bracis-28410 13 5 similar similar ADJ bracis-28410 13 6 content content NOUN bracis-28410 13 7 being be AUX bracis-28410 13 8 viewed view VERB bracis-28410 13 9 by by ADP bracis-28410 13 10 others other NOUN bracis-28410 13 11 compressed compress VERB bracis-28410 13 12 image image NOUN bracis-28410 13 13 super super NOUN bracis-28410 13 14 - - NOUN bracis-28410 13 15 resolution resolution NOUN bracis-28410 13 16 using use VERB bracis-28410 13 17 pre pre ADJ bracis-28410 13 18 - - ADJ bracis-28410 13 19 trained train VERB bracis-28410 13 20 model model NOUN bracis-28410 13 21 assistance assistance NOUN bracis-28410 13 22 chapter chapter NOUN bracis-28410 13 23 © © PROPN bracis-28410 13 24 2023 2023 NUM bracis-28410 13 25 a a DET bracis-28410 13 26 review review NOUN bracis-28410 13 27 of of ADP bracis-28410 13 28 single single ADJ bracis-28410 13 29 image image NOUN bracis-28410 13 30 super super ADJ bracis-28410 13 31 resolution resolution NOUN bracis-28410 13 32 techniques technique NOUN bracis-28410 13 33 using use VERB bracis-28410 13 34 convolutional convolutional ADJ bracis-28410 13 35 neural neural ADJ bracis-28410 13 36 networks network NOUN bracis-28410 13 37 article article NOUN bracis-28410 13 38 13 13 NUM bracis-28410 13 39 september september PROPN bracis-28410 13 40 2023 2023 NUM bracis-28410 13 41 recent recent ADJ bracis-28410 13 42 advances advance NOUN bracis-28410 13 43 in in ADP bracis-28410 13 44 deep deep ADJ bracis-28410 13 45 learning learning NOUN bracis-28410 13 46 for for ADP bracis-28410 13 47 single single ADJ bracis-28410 13 48 image image NOUN bracis-28410 13 49 super super ADJ bracis-28410 13 50 - - ADJ bracis-28410 13 51 resolution resolution ADJ bracis-28410 13 52 chapter chapter NOUN bracis-28410 13 53 © © PROPN bracis-28410 13 54 2018 2018 NUM bracis-28410 13 55 explore explore VERB bracis-28410 13 56 related relate VERB bracis-28410 13 57 subjects subject NOUN bracis-28410 13 58 discover discover VERB bracis-28410 13 59 the the DET bracis-28410 13 60 latest late ADJ bracis-28410 13 61 articles article NOUN bracis-28410 13 62 , , PUNCT bracis-28410 13 63 books book NOUN bracis-28410 13 64 and and CCONJ bracis-28410 13 65 news news NOUN bracis-28410 13 66 in in ADP bracis-28410 13 67 related related ADJ bracis-28410 13 68 subjects subject NOUN bracis-28410 13 69 , , PUNCT bracis-28410 13 70 suggested suggest VERB bracis-28410 13 71 using use VERB bracis-28410 13 72 machine machine NOUN bracis-28410 13 73 learning learning NOUN bracis-28410 13 74 . . PUNCT bracis-28410 14 1 automated automate VERB bracis-28410 14 2 pattern pattern NOUN bracis-28410 14 3 recognition recognition NOUN bracis-28410 14 4 computer computer NOUN bracis-28410 14 5 vision vision NOUN bracis-28410 14 6 supercapacitors supercapacitor NOUN bracis-28410 14 7 super super ADJ bracis-28410 14 8 - - ADJ bracis-28410 14 9 resolution resolution ADJ bracis-28410 14 10 microscopy microscopy NOUN bracis-28410 14 11 video video NOUN bracis-28410 14 12 capsule capsule NOUN bracis-28410 14 13 endoscopy endoscopy VERB bracis-28410 14 14 3 3 NUM bracis-28410 14 15 - - PUNCT bracis-28410 14 16 d d NOUN bracis-28410 14 17 image image NOUN bracis-28410 14 18 reconstruction reconstruction NOUN bracis-28410 14 19 1 1 NUM bracis-28410 14 20 introduction introduction NOUN bracis-28410 14 21 state state NOUN bracis-28410 14 22 - - PUNCT bracis-28410 14 23 of of ADP bracis-28410 14 24 - - PUNCT bracis-28410 14 25 the the DET bracis-28410 14 26 - - PUNCT bracis-28410 14 27 art art NOUN bracis-28410 14 28 methods method NOUN bracis-28410 14 29 in in ADP bracis-28410 14 30 image image NOUN bracis-28410 14 31 super super NOUN bracis-28410 14 32 - - NOUN bracis-28410 14 33 resolution resolution NOUN bracis-28410 14 34 are be AUX bracis-28410 14 35 based base VERB bracis-28410 14 36 on on ADP bracis-28410 14 37 artificial artificial ADJ bracis-28410 14 38 intelligence intelligence NOUN bracis-28410 14 39 concepts concept NOUN bracis-28410 14 40 , , PUNCT bracis-28410 14 41 more more ADV bracis-28410 14 42 specifically specifically ADV bracis-28410 14 43 on on ADP bracis-28410 14 44 deep deep ADJ bracis-28410 14 45 neural neural ADJ bracis-28410 14 46 networks network NOUN bracis-28410 14 47 , , PUNCT bracis-28410 14 48 and and CCONJ bracis-28410 14 49 have have AUX bracis-28410 14 50 achieved achieve VERB bracis-28410 14 51 visually visually ADV bracis-28410 14 52 striking striking ADJ bracis-28410 14 53 results result NOUN bracis-28410 14 54     SPACE bracis-28410 15 1 [ [ X bracis-28410 15 2 6 6 NUM bracis-28410 15 3 , , PUNCT bracis-28410 15 4 22 22 NUM bracis-28410 15 5 , , PUNCT bracis-28410 15 6 40 40 NUM bracis-28410 15 7 , , PUNCT bracis-28410 15 8 61 61 NUM bracis-28410 15 9 , , PUNCT bracis-28410 15 10 63 63 NUM bracis-28410 15 11 ] ] PUNCT bracis-28410 15 12 most most ADJ bracis-28410 15 13 recent recent ADJ bracis-28410 15 14 models model NOUN bracis-28410 15 15 are be AUX bracis-28410 15 16 composed compose VERB bracis-28410 15 17 of of ADP bracis-28410 15 18 traditional traditional ADJ bracis-28410 15 19 convolutional convolutional ADJ bracis-28410 15 20 layers layer NOUN bracis-28410 15 21 that that PRON bracis-28410 15 22 exhibit exhibit VERB bracis-28410 15 23 limitations limitation NOUN bracis-28410 15 24 , , PUNCT bracis-28410 15 25 although although SCONJ bracis-28410 15 26 widely widely ADV bracis-28410 15 27 studied study VERB bracis-28410 15 28 and and CCONJ bracis-28410 15 29 optimized optimize VERB bracis-28410 15 30 . . PUNCT bracis-28410 16 1 for for ADP bracis-28410 16 2 example example NOUN bracis-28410 16 3 , , PUNCT bracis-28410 16 4 they they PRON bracis-28410 16 5 prevent prevent VERB bracis-28410 16 6 a a DET bracis-28410 16 7 better well ADJ bracis-28410 16 8 understanding understanding NOUN bracis-28410 16 9 of of ADP bracis-28410 16 10 the the DET bracis-28410 16 11 data datum NOUN bracis-28410 16 12 by by ADP bracis-28410 16 13 the the DET bracis-28410 16 14 network network NOUN bracis-28410 16 15 , , PUNCT bracis-28410 16 16 not not PART bracis-28410 16 17 absorbing absorb VERB bracis-28410 16 18 valuable valuable ADJ bracis-28410 16 19 information information NOUN bracis-28410 16 20 such such ADJ bracis-28410 16 21 as as ADP bracis-28410 16 22 the the DET bracis-28410 16 23 interrelationship interrelationship NOUN bracis-28410 16 24 between between ADP bracis-28410 16 25 its its PRON bracis-28410 16 26 components component NOUN bracis-28410 16 27 . . PUNCT bracis-28410 17 1 as as ADP bracis-28410 17 2 an an DET bracis-28410 17 3 alternative alternative NOUN bracis-28410 17 4 to to ADP bracis-28410 17 5 convolutional convolutional ADJ bracis-28410 17 6 networks network NOUN bracis-28410 17 7 , , PUNCT bracis-28410 17 8 a a DET bracis-28410 17 9 large large ADJ bracis-28410 17 10 body body NOUN bracis-28410 17 11 of of ADP bracis-28410 17 12 work work NOUN bracis-28410 17 13 has have AUX bracis-28410 17 14 demonstrated demonstrate VERB bracis-28410 17 15 the the DET bracis-28410 17 16 abstraction abstraction ADJ bracis-28410 17 17 power power NOUN bracis-28410 17 18 of of ADP bracis-28410 17 19 the the DET bracis-28410 17 20 capsules capsule NOUN bracis-28410 17 21     SPACE bracis-28410 18 1 [ [ X bracis-28410 18 2 14 14 NUM bracis-28410 18 3 , , PUNCT bracis-28410 18 4 42 42 NUM bracis-28410 18 5 , , PUNCT bracis-28410 18 6 60 60 NUM bracis-28410 18 7 ] ] PUNCT bracis-28410 18 8 . . PUNCT bracis-28410 19 1 for for ADP bracis-28410 19 2 example example NOUN bracis-28410 19 3 , , PUNCT bracis-28410 19 4 sabour sabour PROPN bracis-28410 19 5 et et PROPN bracis-28410 19 6 al al PROPN bracis-28410 19 7 . . PUNCT bracis-28410 19 8     SPACE bracis-28410 20 1 [ [ X bracis-28410 20 2 41 41 NUM bracis-28410 20 3 ] ] PUNCT bracis-28410 20 4 presented present VERB bracis-28410 20 5 an an DET bracis-28410 20 6 implementation implementation NOUN bracis-28410 20 7 of of ADP bracis-28410 20 8 the the DET bracis-28410 20 9 capsule capsule NOUN bracis-28410 20 10 concept concept NOUN bracis-28410 20 11 , , PUNCT bracis-28410 20 12 a a DET bracis-28410 20 13 group group NOUN bracis-28410 20 14 of of ADP bracis-28410 20 15 neurons neuron NOUN bracis-28410 20 16 whose whose DET bracis-28410 20 17 activation activation NOUN bracis-28410 20 18 vector vector NOUN bracis-28410 20 19 represents represent VERB bracis-28410 20 20 the the DET bracis-28410 20 21 parameters parameter NOUN bracis-28410 20 22 that that PRON bracis-28410 20 23 describe describe VERB bracis-28410 20 24 a a DET bracis-28410 20 25 specific specific ADJ bracis-28410 20 26 entity entity NOUN bracis-28410 20 27 , , PUNCT bracis-28410 20 28 such such ADJ bracis-28410 20 29 as as ADP bracis-28410 20 30 an an DET bracis-28410 20 31 object object NOUN bracis-28410 20 32 or or CCONJ bracis-28410 20 33 part part NOUN bracis-28410 20 34 of of ADP bracis-28410 20 35 it it PRON bracis-28410 20 36 , , PUNCT bracis-28410 20 37 and and CCONJ bracis-28410 20 38 its its PRON bracis-28410 20 39 presence presence NOUN bracis-28410 20 40 . . PUNCT bracis-28410 21 1 using use VERB bracis-28410 21 2 a a DET bracis-28410 21 3 three three NUM bracis-28410 21 4 - - PUNCT bracis-28410 21 5 layer layer NOUN bracis-28410 21 6 capsule capsule NOUN bracis-28410 21 7 network network NOUN bracis-28410 21 8 , , PUNCT bracis-28410 21 9 sabour sabour NOUN bracis-28410 21 10 et et PROPN bracis-28410 21 11 al al PROPN bracis-28410 21 12 . . PROPN bracis-28410 21 13 achieved achieve VERB bracis-28410 21 14 results result NOUN bracis-28410 21 15 comparable comparable ADJ bracis-28410 21 16 to to ADP bracis-28410 21 17 those those PRON bracis-28410 21 18 of of ADP bracis-28410 21 19 deeper deep ADJ bracis-28410 21 20 convolutional convolutional ADJ bracis-28410 21 21 neural neural ADJ bracis-28410 21 22 networks network NOUN bracis-28410 21 23 for for ADP bracis-28410 21 24 the the DET bracis-28410 21 25 digit digit NOUN bracis-28410 21 26 identification identification NOUN bracis-28410 21 27 problem problem NOUN bracis-28410 21 28 . . PUNCT bracis-28410 22 1 the the DET bracis-28410 22 2 authors author NOUN bracis-28410 22 3 also also ADV bracis-28410 22 4 obtained obtain VERB bracis-28410 22 5 a a DET bracis-28410 22 6 5 5 NUM bracis-28410 22 7 % % NOUN bracis-28410 22 8 error error NOUN bracis-28410 22 9 rate rate NOUN bracis-28410 22 10 on on ADP bracis-28410 22 11 the the DET bracis-28410 22 12 segmentation segmentation NOUN bracis-28410 22 13 of of ADP bracis-28410 22 14 two two NUM bracis-28410 22 15 different different ADJ bracis-28410 22 16 digits digit NOUN bracis-28410 22 17 with with ADP bracis-28410 22 18 an an DET bracis-28410 22 19 80 80 NUM bracis-28410 22 20 % % NOUN bracis-28410 22 21 overlap overlap NOUN bracis-28410 22 22 – – PUNCT bracis-28410 22 23 a a DET bracis-28410 22 24 rate rate NOUN bracis-28410 22 25 previously previously ADV bracis-28410 22 26 achieved achieve VERB bracis-28410 22 27 only only ADV bracis-28410 22 28 in in ADP bracis-28410 22 29 much much ADV bracis-28410 22 30 simpler simple ADJ bracis-28410 22 31 cases case NOUN bracis-28410 22 32 ( ( PUNCT bracis-28410 22 33 with with ADP bracis-28410 22 34 less less ADJ bracis-28410 22 35 than than ADP bracis-28410 22 36 4 4 NUM bracis-28410 22 37 % % NOUN bracis-28410 22 38 overlap overlap NOUN bracis-28410 22 39 between between ADP bracis-28410 22 40 images image NOUN bracis-28410 22 41 ) ) PUNCT bracis-28410 22 42 . . PUNCT bracis-28410 23 1 hinton hinton PROPN bracis-28410 23 2 et et PROPN bracis-28410 23 3 al al PROPN bracis-28410 23 4 . . PUNCT bracis-28410 23 5     SPACE bracis-28410 24 1 [ [ X bracis-28410 24 2 15 15 NUM bracis-28410 24 3 ] ] PUNCT bracis-28410 24 4 reduced reduce VERB bracis-28410 24 5 by by ADP bracis-28410 24 6 45 45 NUM bracis-28410 24 7 % % NOUN bracis-28410 24 8 the the DET bracis-28410 24 9 error error NOUN bracis-28410 24 10 rate rate NOUN bracis-28410 24 11 on on ADP bracis-28410 24 12 the the DET bracis-28410 24 13 problem problem NOUN bracis-28410 24 14 of of ADP bracis-28410 24 15 identifying identify VERB bracis-28410 24 16 objects object NOUN bracis-28410 24 17 from from ADP bracis-28410 24 18 different different ADJ bracis-28410 24 19 perspectives perspective NOUN bracis-28410 24 20 while while SCONJ bracis-28410 24 21 making make VERB bracis-28410 24 22 the the DET bracis-28410 24 23 network network NOUN bracis-28410 24 24 more more ADV bracis-28410 24 25 resistant resistant ADJ bracis-28410 24 26 to to ADP bracis-28410 24 27 adversarial adversarial ADJ bracis-28410 24 28 attacks attack NOUN bracis-28410 24 29 . . PUNCT bracis-28410 25 1 lalonde lalonde NOUN bracis-28410 25 2 and and CCONJ bracis-28410 25 3 bagci bagci NOUN bracis-28410 25 4     SPACE bracis-28410 26 1 [ [ X bracis-28410 26 2 28 28 NUM bracis-28410 26 3 ] ] PUNCT bracis-28410 26 4 have have AUX bracis-28410 26 5 created create VERB bracis-28410 26 6 a a DET bracis-28410 26 7 network network NOUN bracis-28410 26 8 able able ADJ bracis-28410 26 9 to to PART bracis-28410 26 10 process process VERB bracis-28410 26 11 images image NOUN bracis-28410 26 12 tens ten NOUN bracis-28410 26 13 of of ADP bracis-28410 26 14 times time NOUN bracis-28410 26 15 larger large ADJ bracis-28410 26 16 while while SCONJ bracis-28410 26 17 reducing reduce VERB bracis-28410 26 18 the the DET bracis-28410 26 19 number number NOUN bracis-28410 26 20 of of ADP bracis-28410 26 21 parameters parameter NOUN bracis-28410 26 22 by by ADP bracis-28410 26 23 38.4 38.4 NUM bracis-28410 26 24 % % NOUN bracis-28410 26 25 and and CCONJ bracis-28410 26 26 increasing increase VERB bracis-28410 26 27 the the DET bracis-28410 26 28 quality quality NOUN bracis-28410 26 29 of of ADP bracis-28410 26 30 medical medical ADJ bracis-28410 26 31 image image NOUN bracis-28410 26 32 segmentation segmentation NOUN bracis-28410 26 33 . . PUNCT bracis-28410 27 1 a a DET bracis-28410 27 2 few few ADJ bracis-28410 27 3 works work NOUN bracis-28410 27 4 proposed propose VERB bracis-28410 27 5 the the DET bracis-28410 27 6 usage usage NOUN bracis-28410 27 7 of of ADP bracis-28410 27 8 capsule capsule NOUN bracis-28410 27 9 - - PUNCT bracis-28410 27 10 based base VERB bracis-28410 27 11 networks network NOUN bracis-28410 27 12 to to PART bracis-28410 27 13 solve solve VERB bracis-28410 27 14 problems problem NOUN bracis-28410 27 15 that that PRON bracis-28410 27 16 involve involve VERB bracis-28410 27 17 low low ADJ bracis-28410 27 18 - - PUNCT bracis-28410 27 19 resolution resolution NOUN bracis-28410 27 20 images image NOUN bracis-28410 27 21 . . PUNCT bracis-28410 28 1 singh singh PROPN bracis-28410 28 2 et et PROPN bracis-28410 28 3 al al PROPN bracis-28410 28 4 . . PUNCT bracis-28410 28 5     SPACE bracis-28410 29 1 [ [ X bracis-28410 29 2 46 46 NUM bracis-28410 29 3 ] ] PUNCT bracis-28410 29 4 proposed propose VERB bracis-28410 29 5 a a DET bracis-28410 29 6 dual dual ADJ bracis-28410 29 7 directed direct VERB bracis-28410 29 8 capsule capsule NOUN bracis-28410 29 9 network network NOUN bracis-28410 29 10 , , PUNCT bracis-28410 29 11 named name VERB bracis-28410 29 12 directcapsnet directcapsnet NOUN bracis-28410 29 13 , , PUNCT bracis-28410 29 14 which which PRON bracis-28410 29 15 employs employ VERB bracis-28410 29 16 a a DET bracis-28410 29 17 combination combination NOUN bracis-28410 29 18 of of ADP bracis-28410 29 19 capsules capsule NOUN bracis-28410 29 20 and and CCONJ bracis-28410 29 21 convolutional convolutional ADJ bracis-28410 29 22 layers layer NOUN bracis-28410 29 23 for for ADP bracis-28410 29 24 addressing address VERB bracis-28410 29 25 very very ADV bracis-28410 29 26 low low ADJ bracis-28410 29 27 resolution resolution NOUN bracis-28410 29 28 ( ( PUNCT bracis-28410 29 29 vlr vlr ADJ bracis-28410 29 30 ) ) PUNCT bracis-28410 29 31 digit digit NOUN bracis-28410 29 32 and and CCONJ bracis-28410 29 33 face face VERB bracis-28410 29 34 recognition recognition NOUN bracis-28410 29 35 problems problem NOUN bracis-28410 29 36 . . PUNCT bracis-28410 30 1 majdabadi majdabadi NOUN bracis-28410 30 2 and and CCONJ bracis-28410 30 3 ko ko PROPN bracis-28410 30 4     SPACE bracis-28410 31 1 [ [ X bracis-28410 31 2 35 35 NUM bracis-28410 31 3 ] ] PUNCT bracis-28410 31 4 implemented implement VERB bracis-28410 31 5 a a DET bracis-28410 31 6 generative generative ADJ bracis-28410 31 7 adversarial adversarial ADJ bracis-28410 31 8 network network NOUN bracis-28410 31 9 ( ( PUNCT bracis-28410 31 10 gan gan PROPN bracis-28410 31 11 ) ) PUNCT bracis-28410 31 12 that that PRON bracis-28410 31 13 uses use VERB bracis-28410 31 14 a a DET bracis-28410 31 15 capsnet capsnet NOUN bracis-28410 31 16 as as ADP bracis-28410 31 17 the the DET bracis-28410 31 18 discriminator discriminator NOUN bracis-28410 31 19 for for ADP bracis-28410 31 20 facial facial ADJ bracis-28410 31 21 image image NOUN bracis-28410 31 22 super super NOUN bracis-28410 31 23 - - NOUN bracis-28410 31 24 resolution resolution NOUN bracis-28410 31 25 , , PUNCT bracis-28410 31 26 surpassing surpass VERB bracis-28410 31 27 strong strong ADJ bracis-28410 31 28 baselines baseline NOUN bracis-28410 31 29 in in ADP bracis-28410 31 30 all all DET bracis-28410 31 31 metrics metric NOUN bracis-28410 31 32 . . PUNCT bracis-28410 32 1 hsu hsu PROPN bracis-28410 32 2 et et PROPN bracis-28410 32 3 al al PROPN bracis-28410 32 4 . . PUNCT bracis-28410 32 5     SPACE bracis-28410 33 1 [ [ X bracis-28410 33 2 18 18 NUM bracis-28410 33 3 ] ] PUNCT bracis-28410 33 4     SPACE bracis-28410 33 5 developed develop VERB bracis-28410 33 6 two two NUM bracis-28410 33 7 frameworks framework NOUN bracis-28410 33 8 to to PART bracis-28410 33 9 incorporate incorporate VERB bracis-28410 33 10 capsules capsule NOUN bracis-28410 33 11 into into ADP bracis-28410 33 12 image image NOUN bracis-28410 33 13 sr sr PROPN bracis-28410 33 14 convolutional convolutional ADJ bracis-28410 33 15 networks network NOUN bracis-28410 33 16 : : PUNCT bracis-28410 33 17 capsule capsule NOUN bracis-28410 33 18 image image NOUN bracis-28410 33 19 restoration restoration NOUN bracis-28410 33 20 neural neural ADJ bracis-28410 33 21 network network NOUN bracis-28410 33 22 ( ( PUNCT bracis-28410 33 23 cirnn cirnn NOUN bracis-28410 33 24 ) ) PUNCT bracis-28410 33 25 and and CCONJ bracis-28410 33 26 capsule capsule ADJ bracis-28410 33 27 attention attention NOUN bracis-28410 33 28 and and CCONJ bracis-28410 33 29 reconstruction reconstruction NOUN bracis-28410 33 30 neural neural ADJ bracis-28410 33 31 network network NOUN bracis-28410 33 32 ( ( PUNCT bracis-28410 33 33 carnn carnn PROPN bracis-28410 33 34 ) ) PUNCT bracis-28410 33 35 . . PUNCT bracis-28410 34 1 the the DET bracis-28410 34 2 results result NOUN bracis-28410 34 3 outperformed outperform VERB bracis-28410 34 4 traditional traditional ADJ bracis-28410 34 5 cnn cnn PROPN bracis-28410 34 6 methods method NOUN bracis-28410 34 7 with with ADP bracis-28410 34 8 a a DET bracis-28410 34 9 similar similar ADJ bracis-28410 34 10 number number NOUN bracis-28410 34 11 of of ADP bracis-28410 34 12 parameters parameter NOUN bracis-28410 34 13 . . PUNCT bracis-28410 35 1 despite despite SCONJ bracis-28410 35 2 the the DET bracis-28410 35 3 positive positive ADJ bracis-28410 35 4 results result NOUN bracis-28410 35 5 , , PUNCT bracis-28410 35 6 most most ADJ bracis-28410 35 7 of of ADP bracis-28410 35 8 these these DET bracis-28410 35 9 works work NOUN bracis-28410 35 10 rely rely VERB bracis-28410 35 11 on on ADP bracis-28410 35 12 plain plain ADJ bracis-28410 35 13 capsnet capsnet NOUN bracis-28410 35 14     SPACE bracis-28410 36 1 [ [ X bracis-28410 36 2 41 41 NUM bracis-28410 36 3 ] ] PUNCT bracis-28410 36 4 , , PUNCT bracis-28410 36 5 failing fail VERB bracis-28410 36 6 to to PART bracis-28410 36 7 explore explore VERB bracis-28410 36 8 novel novel ADJ bracis-28410 36 9 capsule capsule NOUN bracis-28410 36 10 concepts concept NOUN bracis-28410 36 11 . . PUNCT bracis-28410 37 1 our our PRON bracis-28410 37 2 method method NOUN bracis-28410 37 3 bridges bridge NOUN bracis-28410 37 4 this this DET bracis-28410 37 5 gap gap NOUN bracis-28410 37 6 . . PUNCT bracis-28410 38 1 this this DET bracis-28410 38 2 work work NOUN bracis-28410 38 3 focuses focus VERB bracis-28410 38 4 on on ADP bracis-28410 38 5 the the DET bracis-28410 38 6 single single ADJ bracis-28410 38 7 image image NOUN bracis-28410 38 8 super super ADJ bracis-28410 38 9 - - ADJ bracis-28410 38 10 resolution resolution ADJ bracis-28410 38 11 problem problem NOUN bracis-28410 38 12 using use VERB bracis-28410 38 13 capsules capsule NOUN bracis-28410 38 14 . . PUNCT bracis-28410 39 1 we we PRON bracis-28410 39 2 implement implement VERB bracis-28410 39 3 a a DET bracis-28410 39 4 model model NOUN bracis-28410 39 5 based base VERB bracis-28410 39 6 on on ADP bracis-28410 39 7 newer new ADJ bracis-28410 39 8 concepts concept NOUN bracis-28410 39 9 of of ADP bracis-28410 39 10 capsules capsule NOUN bracis-28410 39 11 for for ADP bracis-28410 39 12 single single ADJ bracis-28410 39 13 image image NOUN bracis-28410 39 14 super super NOUN bracis-28410 39 15 - - NOUN bracis-28410 39 16 resolution resolution ADJ bracis-28410 39 17 ( ( PUNCT bracis-28410 39 18 sisr sisr NOUN bracis-28410 39 19 ) ) PUNCT bracis-28410 39 20 problems problem NOUN bracis-28410 39 21 . . PUNCT bracis-28410 40 1 throughout throughout ADP bracis-28410 40 2 our our PRON bracis-28410 40 3 evaluation evaluation NOUN bracis-28410 40 4 , , PUNCT bracis-28410 40 5 we we PRON bracis-28410 40 6 used use VERB bracis-28410 40 7 publicly publicly ADV bracis-28410 40 8 available available ADJ bracis-28410 40 9 and and CCONJ bracis-28410 40 10 standard standard ADJ bracis-28410 40 11 benchmarks benchmark NOUN bracis-28410 40 12 such such ADJ bracis-28410 40 13 as as ADP bracis-28410 40 14 set5 set5 PROPN bracis-28410 40 15     SPACE bracis-28410 41 1 [ [ X bracis-28410 41 2 7 7 NUM bracis-28410 41 3 ] ] PUNCT bracis-28410 41 4 , , PUNCT bracis-28410 41 5 set14 set14 NOUN bracis-28410 41 6     SPACE bracis-28410 42 1 [ [ X bracis-28410 42 2 59 59 NUM bracis-28410 42 3 ] ] PUNCT bracis-28410 42 4 , , PUNCT bracis-28410 42 5 b100 b100 PROPN bracis-28410 42 6     SPACE bracis-28410 43 1 [ [ X bracis-28410 43 2 36 36 NUM bracis-28410 43 3 ] ] PUNCT bracis-28410 43 4 , , PUNCT bracis-28410 43 5 urban100 urban100 PROPN bracis-28410 43 6     SPACE bracis-28410 44 1 [ [ X bracis-28410 44 2 19 19 NUM bracis-28410 44 3 ] ] PUNCT bracis-28410 44 4 and and CCONJ bracis-28410 44 5 div2k div2k PROPN bracis-28410 44 6     SPACE bracis-28410 44 7 [ [ X bracis-28410 44 8 2 2 NUM bracis-28410 44 9 ] ] PUNCT bracis-28410 44 10 . . PUNCT bracis-28410 45 1 our our PRON bracis-28410 45 2 model model NOUN bracis-28410 45 3 yields yield VERB bracis-28410 45 4 fine fine ADV bracis-28410 45 5 - - PUNCT bracis-28410 45 6 grained grain VERB bracis-28410 45 7 super super ADJ bracis-28410 45 8 - - ADJ bracis-28410 45 9 resolution resolution ADJ bracis-28410 45 10 images image NOUN bracis-28410 45 11 and and CCONJ bracis-28410 45 12 achieves achieve VERB bracis-28410 45 13 positive positive ADJ bracis-28410 45 14 results result NOUN bracis-28410 45 15 with with ADP bracis-28410 45 16 fewer few ADJ bracis-28410 45 17 convolutional convolutional ADJ bracis-28410 45 18 - - PUNCT bracis-28410 45 19 based base VERB bracis-28410 45 20 layers layer NOUN bracis-28410 45 21 than than ADP bracis-28410 45 22 baselines baseline NOUN bracis-28410 45 23 . . PUNCT bracis-28410 46 1 2 2 NUM bracis-28410 46 2 background background NOUN bracis-28410 46 3 super super NOUN bracis-28410 46 4 - - NOUN bracis-28410 46 5 resolution resolution NOUN bracis-28410 46 6 . . PUNCT bracis-28410 47 1 super super ADJ bracis-28410 47 2 - - NOUN bracis-28410 47 3 resolution resolution NOUN bracis-28410 47 4 ( ( PUNCT bracis-28410 47 5 sr sr PROPN bracis-28410 47 6 ) ) PUNCT bracis-28410 47 7 is be AUX bracis-28410 47 8 the the DET bracis-28410 47 9 process process NOUN bracis-28410 47 10 of of ADP bracis-28410 47 11 obtaining obtain VERB bracis-28410 47 12 one one NUM bracis-28410 47 13 or or CCONJ bracis-28410 47 14 more more ADV bracis-28410 47 15 plausible plausible ADJ bracis-28410 47 16 high high ADJ bracis-28410 47 17 - - PUNCT bracis-28410 47 18 resolution resolution NOUN bracis-28410 47 19 images image NOUN bracis-28410 47 20 ( ( PUNCT bracis-28410 47 21 hr hr NOUN bracis-28410 47 22 ) ) PUNCT bracis-28410 47 23 from from ADP bracis-28410 47 24 one one NUM bracis-28410 47 25 or or CCONJ bracis-28410 47 26 more more ADV bracis-28410 47 27 low low ADJ bracis-28410 47 28 - - PUNCT bracis-28410 47 29 resolution resolution NOUN bracis-28410 47 30 images image NOUN bracis-28410 47 31 ( ( PUNCT bracis-28410 47 32 lr lr NOUN bracis-28410 47 33 ) ) PUNCT bracis-28410 47 34     SPACE bracis-28410 48 1 [ [ X bracis-28410 48 2 4 4 NUM bracis-28410 48 3 , , PUNCT bracis-28410 48 4 37 37 NUM bracis-28410 48 5 , , PUNCT bracis-28410 48 6 56 56 NUM bracis-28410 48 7 ] ] PUNCT bracis-28410 48 8 . . PUNCT bracis-28410 49 1 it it PRON bracis-28410 49 2 is be AUX bracis-28410 49 3 an an DET bracis-28410 49 4 area area NOUN bracis-28410 49 5 that that PRON bracis-28410 49 6 has have AUX bracis-28410 49 7 been be AUX bracis-28410 49 8 studied study VERB bracis-28410 49 9 for for ADP bracis-28410 49 10 decades decade NOUN bracis-28410 49 11     SPACE bracis-28410 50 1 [ [ X bracis-28410 50 2 21 21 NUM bracis-28410 50 3 ] ] PUNCT bracis-28410 50 4 and and CCONJ bracis-28410 50 5 has have VERB bracis-28410 50 6 a a DET bracis-28410 50 7 wide wide ADJ bracis-28410 50 8 variety variety NOUN bracis-28410 50 9 of of ADP bracis-28410 50 10 application application NOUN bracis-28410 50 11 fields field NOUN bracis-28410 50 12 such such ADJ bracis-28410 50 13 as as ADP bracis-28410 50 14 smart smart ADJ bracis-28410 50 15 surveillance surveillance NOUN bracis-28410 50 16 , , PUNCT bracis-28410 50 17 aerial aerial ADJ bracis-28410 50 18 imaging imaging NOUN bracis-28410 50 19 , , PUNCT bracis-28410 50 20 medical medical ADJ bracis-28410 50 21 image image NOUN bracis-28410 50 22 processing processing NOUN bracis-28410 50 23 , , PUNCT bracis-28410 50 24 and and CCONJ bracis-28410 50 25 traffic traffic NOUN bracis-28410 50 26 sign sign NOUN bracis-28410 50 27 reading read VERB bracis-28410 50 28     SPACE bracis-28410 51 1 [ [ X bracis-28410 51 2 4 4 NUM bracis-28410 51 3 , , PUNCT bracis-28410 51 4 37 37 NUM bracis-28410 51 5 , , PUNCT bracis-28410 51 6 44 44 NUM bracis-28410 51 7 , , PUNCT bracis-28410 51 8 56 56 NUM bracis-28410 51 9 ] ] PUNCT bracis-28410 51 10 . . PUNCT bracis-28410 52 1 the the DET bracis-28410 52 2 relationship relationship NOUN bracis-28410 52 3 between between ADP bracis-28410 52 4 lr lr NOUN bracis-28410 52 5 and and CCONJ bracis-28410 52 6 hr hr NOUN bracis-28410 52 7 images image NOUN bracis-28410 52 8 may may AUX bracis-28410 52 9 vary vary VERB bracis-28410 52 10 depending depend VERB bracis-28410 52 11 on on ADP bracis-28410 52 12 the the DET bracis-28410 52 13 situation situation NOUN bracis-28410 52 14 . . PUNCT bracis-28410 53 1 many many ADJ bracis-28410 53 2 studies study NOUN bracis-28410 53 3 assume assume VERB bracis-28410 53 4 that that SCONJ bracis-28410 53 5 the the DET bracis-28410 53 6 lr lr NOUN bracis-28410 53 7 image image NOUN bracis-28410 53 8 is be AUX bracis-28410 53 9 a a DET bracis-28410 53 10 reduced reduced ADJ bracis-28410 53 11 version version NOUN bracis-28410 53 12 of of ADP bracis-28410 53 13 the the DET bracis-28410 53 14 hr hr NOUN bracis-28410 53 15 image image NOUN bracis-28410 53 16 by by ADP bracis-28410 53 17 bicubic bicubic ADJ bracis-28410 53 18 interpolation interpolation NOUN bracis-28410 53 19 . . PUNCT bracis-28410 54 1 however however ADV bracis-28410 54 2 , , PUNCT bracis-28410 54 3 other other ADJ bracis-28410 54 4 degradation degradation NOUN bracis-28410 54 5 factors factor NOUN bracis-28410 54 6 can can AUX bracis-28410 54 7 be be AUX bracis-28410 54 8 considered consider VERB bracis-28410 54 9 in in ADP bracis-28410 54 10 real real ADJ bracis-28410 54 11 examples example NOUN bracis-28410 54 12 , , PUNCT bracis-28410 54 13 such such ADJ bracis-28410 54 14 as as ADP bracis-28410 54 15 quantization quantization NOUN bracis-28410 54 16 errors error NOUN bracis-28410 54 17 , , PUNCT bracis-28410 54 18 acquisition acquisition NOUN bracis-28410 54 19 sensor sensor NOUN bracis-28410 54 20 limitations limitation NOUN bracis-28410 54 21 , , PUNCT bracis-28410 54 22 presence presence NOUN bracis-28410 54 23 of of ADP bracis-28410 54 24 noise noise NOUN bracis-28410 54 25 , , PUNCT bracis-28410 54 26 blurring blurring NOUN bracis-28410 54 27 , , PUNCT bracis-28410 54 28 and and CCONJ bracis-28410 54 29 even even ADV bracis-28410 54 30 the the DET bracis-28410 54 31 use use NOUN bracis-28410 54 32 of of ADP bracis-28410 54 33 different different ADJ bracis-28410 54 34 interpolation interpolation NOUN bracis-28410 54 35 operators operator NOUN bracis-28410 54 36 aiming aim VERB bracis-28410 54 37 resolution resolution NOUN bracis-28410 54 38 reduction reduction NOUN bracis-28410 54 39 for for ADP bracis-28410 54 40 storage storage NOUN bracis-28410 54 41     SPACE bracis-28410 55 1 [ [ X bracis-28410 55 2 50 50 NUM bracis-28410 55 3 ] ] PUNCT bracis-28410 55 4 . . PUNCT bracis-28410 56 1 the the DET bracis-28410 56 2 first first ADJ bracis-28410 56 3 successful successful ADJ bracis-28410 56 4 usage usage NOUN bracis-28410 56 5 of of ADP bracis-28410 56 6 neural neural ADJ bracis-28410 56 7 networks network NOUN bracis-28410 56 8 for for ADP bracis-28410 56 9 sisr sisr NOUN bracis-28410 56 10 problems problem NOUN bracis-28410 56 11 was be AUX bracis-28410 56 12 developed develop VERB bracis-28410 56 13 by by ADP bracis-28410 56 14 dong dong PROPN bracis-28410 56 15 et et PROPN bracis-28410 56 16 al al PROPN bracis-28410 56 17 . . PUNCT bracis-28410 56 18     SPACE bracis-28410 57 1 [ [ X bracis-28410 57 2 11 11 NUM bracis-28410 57 3 ] ] PUNCT bracis-28410 57 4 . . PUNCT bracis-28410 58 1 with with ADP bracis-28410 58 2 the the DET bracis-28410 58 3 super super ADJ bracis-28410 58 4 - - ADJ bracis-28410 58 5 resolution resolution ADJ bracis-28410 58 6 convolutional convolutional ADJ bracis-28410 58 7 neural neural ADJ bracis-28410 58 8 network network NOUN bracis-28410 58 9 ( ( PUNCT bracis-28410 58 10 srcnn srcnn NOUN bracis-28410 58 11 ) ) PUNCT bracis-28410 58 12 model model NOUN bracis-28410 58 13 , , PUNCT bracis-28410 58 14 they they PRON bracis-28410 58 15 created create VERB bracis-28410 58 16 a a DET bracis-28410 58 17 complete complete ADJ bracis-28410 58 18 solution solution NOUN bracis-28410 58 19 that that PRON bracis-28410 58 20 maps map VERB bracis-28410 58 21 lr lr NOUN bracis-28410 58 22 images image NOUN bracis-28410 58 23 to to ADP bracis-28410 58 24 sr sr PROPN bracis-28410 58 25 versions version NOUN bracis-28410 58 26 with with ADP bracis-28410 58 27 little little ADJ bracis-28410 58 28 pre pre ADJ bracis-28410 58 29 / / SYM bracis-28410 58 30 post post ADJ bracis-28410 58 31 - - ADJ bracis-28410 58 32 processing processing NOUN bracis-28410 58 33 , , PUNCT bracis-28410 58 34 yielding yield VERB bracis-28410 58 35 superior superior ADJ bracis-28410 58 36 results result NOUN bracis-28410 58 37 . . PUNCT bracis-28410 59 1 after after ADP bracis-28410 59 2 their their PRON bracis-28410 59 3 achievement achievement NOUN bracis-28410 59 4 , , PUNCT bracis-28410 59 5 several several ADJ bracis-28410 59 6 other other ADJ bracis-28410 59 7 works work NOUN bracis-28410 59 8 have have AUX bracis-28410 59 9 advanced advance VERB bracis-28410 59 10 the the DET bracis-28410 59 11 state state NOUN bracis-28410 59 12 - - PUNCT bracis-28410 59 13 of of ADP bracis-28410 59 14 - - PUNCT bracis-28410 59 15 the the DET bracis-28410 59 16 - - PUNCT bracis-28410 59 17 art art NOUN bracis-28410 59 18 in in ADP bracis-28410 59 19 the the DET bracis-28410 59 20 sisr sisr PROPN bracis-28410 59 21 problem problem NOUN bracis-28410 59 22     SPACE bracis-28410 59 23 [ [ X bracis-28410 59 24 10 10 NUM bracis-28410 59 25 , , PUNCT bracis-28410 59 26 24 24 NUM bracis-28410 59 27 , , PUNCT bracis-28410 59 28 25 25 NUM bracis-28410 59 29 , , PUNCT bracis-28410 59 30 54 54 NUM bracis-28410 59 31 ] ] PUNCT bracis-28410 59 32 , , PUNCT bracis-28410 59 33 but but CCONJ bracis-28410 59 34 they they PRON bracis-28410 59 35 have have VERB bracis-28410 59 36 strong strong ADJ bracis-28410 59 37 limitations limitation NOUN bracis-28410 59 38 . . PUNCT bracis-28410 60 1 these these DET bracis-28410 60 2 models model NOUN bracis-28410 60 3 receive receive VERB bracis-28410 60 4 as as ADP bracis-28410 60 5 input input NOUN bracis-28410 60 6 an an DET bracis-28410 60 7 enlarged enlarged ADJ bracis-28410 60 8 version version NOUN bracis-28410 60 9 of of ADP bracis-28410 60 10 the the DET bracis-28410 60 11 lr lr NOUN bracis-28410 60 12 image image NOUN bracis-28410 60 13 , , PUNCT bracis-28410 60 14 usually usually ADV bracis-28410 60 15 through through ADP bracis-28410 60 16 bicubic bicubic ADJ bracis-28410 60 17 interpolation interpolation NOUN bracis-28410 60 18 , , PUNCT bracis-28410 60 19 and and CCONJ bracis-28410 60 20 seek seek VERB bracis-28410 60 21 to to PART bracis-28410 60 22 improve improve VERB bracis-28410 60 23 the the DET bracis-28410 60 24 quality quality NOUN bracis-28410 60 25 of of ADP bracis-28410 60 26 the the DET bracis-28410 60 27 image image NOUN bracis-28410 60 28 . . PUNCT bracis-28410 61 1 this this PRON bracis-28410 61 2 means mean VERB bracis-28410 61 3 that that SCONJ bracis-28410 61 4 the the DET bracis-28410 61 5 operations operation NOUN bracis-28410 61 6 performed perform VERB bracis-28410 61 7 by by ADP bracis-28410 61 8 the the DET bracis-28410 61 9 neural neural ADJ bracis-28410 61 10 network network NOUN bracis-28410 61 11 are be AUX bracis-28410 61 12 all all ADV bracis-28410 61 13 done do VERB bracis-28410 61 14 in in ADP bracis-28410 61 15 high high ADJ bracis-28410 61 16 - - PUNCT bracis-28410 61 17 resolution resolution NOUN bracis-28410 61 18 space space NOUN bracis-28410 61 19 , , PUNCT bracis-28410 61 20 which which PRON bracis-28410 61 21 is be AUX bracis-28410 61 22 inefficient inefficient ADJ bracis-28410 61 23 and and CCONJ bracis-28410 61 24 incurs incur VERB bracis-28410 61 25 high high ADJ bracis-28410 61 26 processing processing NOUN bracis-28410 61 27 cost cost NOUN bracis-28410 61 28 . . PUNCT bracis-28410 62 1 the the DET bracis-28410 62 2 computational computational ADJ bracis-28410 62 3 complexity complexity NOUN bracis-28410 62 4 of of ADP bracis-28410 62 5 the the DET bracis-28410 62 6 convolution convolution NOUN bracis-28410 62 7 grows grow VERB bracis-28410 62 8 quadratically quadratically ADV bracis-28410 62 9 with with ADP bracis-28410 62 10 the the DET bracis-28410 62 11 size size NOUN bracis-28410 62 12 of of ADP bracis-28410 62 13 the the DET bracis-28410 62 14 input input NOUN bracis-28410 62 15 image image NOUN bracis-28410 62 16 , , PUNCT bracis-28410 62 17 whose whose DET bracis-28410 62 18 generation generation NOUN bracis-28410 62 19 of of ADP bracis-28410 62 20 an an DET bracis-28410 62 21 sr sr PROPN bracis-28410 62 22 image image NOUN bracis-28410 62 23 with with ADP bracis-28410 62 24 a a DET bracis-28410 62 25 scaling scale VERB bracis-28410 62 26 factor factor NOUN bracis-28410 62 27 n n PRON bracis-28410 62 28 would would AUX bracis-28410 62 29 result result VERB bracis-28410 62 30 in in ADP bracis-28410 62 31 a a DET bracis-28410 62 32 cost cost NOUN bracis-28410 62 33 \(n^{2}\ \(n^{2}\ NOUN bracis-28410 62 34 ) ) PUNCT bracis-28410 62 35 compared compare VERB bracis-28410 62 36 to to ADP bracis-28410 62 37 the the DET bracis-28410 62 38 processing processing NOUN bracis-28410 62 39 in in ADP bracis-28410 62 40 the the DET bracis-28410 62 41 low low ADJ bracis-28410 62 42 - - PUNCT bracis-28410 62 43 resolution resolution NOUN bracis-28410 62 44 space space NOUN bracis-28410 62 45     SPACE bracis-28410 63 1 [ [ X bracis-28410 63 2 12 12 NUM bracis-28410 63 3 ] ] PUNCT bracis-28410 63 4 . . PUNCT bracis-28410 64 1 looking look VERB bracis-28410 64 2 for for ADP bracis-28410 64 3 a a DET bracis-28410 64 4 way way NOUN bracis-28410 64 5 of of ADP bracis-28410 64 6 postponing postpone VERB bracis-28410 64 7 the the DET bracis-28410 64 8 resolution resolution NOUN bracis-28410 64 9 increase increase NOUN bracis-28410 64 10 in in ADP bracis-28410 64 11 the the DET bracis-28410 64 12 network network NOUN bracis-28410 64 13 , , PUNCT bracis-28410 64 14 shi shi PROPN bracis-28410 64 15 et et PROPN bracis-28410 64 16 al al PROPN bracis-28410 64 17 . . PUNCT bracis-28410 64 18     SPACE bracis-28410 65 1 [ [ X bracis-28410 65 2 44 44 NUM bracis-28410 65 3 ] ] PUNCT bracis-28410 65 4 developed develop VERB bracis-28410 65 5 a a DET bracis-28410 65 6 new new ADJ bracis-28410 65 7 layer layer NOUN bracis-28410 65 8 , , PUNCT bracis-28410 65 9 called call VERB bracis-28410 65 10 the the DET bracis-28410 65 11 subpixel subpixel ADJ bracis-28410 65 12 convolution convolution NOUN bracis-28410 65 13 ( ( PUNCT bracis-28410 65 14 or or CCONJ bracis-28410 65 15 pixelshuffle pixelshuffle ADV bracis-28410 65 16 ) ) PUNCT bracis-28410 65 17 . . PUNCT bracis-28410 66 1 such such DET bracis-28410 66 2 a a DET bracis-28410 66 3 layer layer NOUN bracis-28410 66 4 works work VERB bracis-28410 66 5 equivalent equivalent ADJ bracis-28410 66 6 to to AUX bracis-28410 66 7 deconvolution deconvolution VERB bracis-28410 66 8 with with ADP bracis-28410 66 9 kernel kernel PROPN bracis-28410 66 10 size size NOUN bracis-28410 66 11 divisible divisible ADJ bracis-28410 66 12 by by ADP bracis-28410 66 13 spacing space VERB bracis-28410 66 14 , , PUNCT bracis-28410 66 15 but but CCONJ bracis-28410 66 16 it it PRON bracis-28410 66 17 is be AUX bracis-28410 66 18 \(\log \(\log NOUN bracis-28410 67 1 _ _ PRON bracis-28410 67 2 { { PUNCT bracis-28410 67 3 2}r^{2}\ 2}r^{2}\ NUM bracis-28410 67 4 ) ) PUNCT bracis-28410 67 5 times time NOUN bracis-28410 67 6 faster fast ADV bracis-28410 67 7 than than ADP bracis-28410 67 8 deconvolution deconvolution NOUN bracis-28410 67 9 . . PUNCT bracis-28410 68 1 their their PRON bracis-28410 68 2 network network NOUN bracis-28410 68 3 , , PUNCT bracis-28410 68 4 named name VERB bracis-28410 68 5 efficient efficient ADJ bracis-28410 68 6 sub sub ADJ bracis-28410 68 7 - - ADJ bracis-28410 68 8 pixel pixel ADJ bracis-28410 68 9 convolutional convolutional ADJ bracis-28410 68 10 neural neural ADJ bracis-28410 68 11 network network NOUN bracis-28410 68 12 ( ( PUNCT bracis-28410 68 13 espcn espcn PROPN bracis-28410 68 14 ) ) PUNCT bracis-28410 68 15 , , PUNCT bracis-28410 68 16 achieved achieve VERB bracis-28410 68 17 speed speed NOUN bracis-28410 68 18 improvements improvement NOUN bracis-28410 68 19 of of ADP bracis-28410 68 20 over over ADP bracis-28410 68 21 \(10\times \(10\time NOUN bracis-28410 68 22 \ \ NOUN bracis-28410 68 23 ) ) PUNCT bracis-28410 68 24 compared compare VERB bracis-28410 68 25 to to ADP bracis-28410 68 26 srcnn srcnn PROPN bracis-28410 68 27     SPACE bracis-28410 69 1 [ [ X bracis-28410 69 2 11 11 NUM bracis-28410 69 3 ] ] PUNCT bracis-28410 69 4 while while SCONJ bracis-28410 69 5 having have VERB bracis-28410 69 6 a a DET bracis-28410 69 7 higher high ADJ bracis-28410 69 8 number number NOUN bracis-28410 69 9 of of ADP bracis-28410 69 10 parameters parameter NOUN bracis-28410 69 11 and and CCONJ bracis-28410 69 12 achieving achieve VERB bracis-28410 69 13 better well ADJ bracis-28410 69 14 results result NOUN bracis-28410 69 15 for for ADP bracis-28410 69 16 an an DET bracis-28410 69 17 upscaling upscale VERB bracis-28410 69 18 factor factor NOUN bracis-28410 69 19 of of ADP bracis-28410 69 20 \(4\times \(4\times PROPN bracis-28410 69 21 \ \ NOUN bracis-28410 69 22 ) ) PUNCT bracis-28410 69 23 . . PUNCT bracis-28410 70 1 the the DET bracis-28410 70 2 concept concept NOUN bracis-28410 70 3 of of ADP bracis-28410 70 4 subpixel subpixel ADJ bracis-28410 70 5 convolution convolution NOUN bracis-28410 70 6 is be AUX bracis-28410 70 7 currently currently ADV bracis-28410 70 8 a a DET bracis-28410 70 9 common common ADJ bracis-28410 70 10 choice choice NOUN bracis-28410 70 11 to to PART bracis-28410 70 12 perform perform VERB bracis-28410 70 13 upscaling upscale VERB bracis-28410 70 14 in in ADP bracis-28410 70 15 neural neural ADJ bracis-28410 70 16 networks network NOUN bracis-28410 70 17 . . PUNCT bracis-28410 71 1 it it PRON bracis-28410 71 2 has have AUX bracis-28410 71 3 been be AUX bracis-28410 71 4 used use VERB bracis-28410 71 5 by by ADP bracis-28410 71 6 several several ADJ bracis-28410 71 7 solutions solution NOUN bracis-28410 71 8 that that PRON bracis-28410 71 9 have have AUX bracis-28410 71 10 reached reach VERB bracis-28410 71 11 the the DET bracis-28410 71 12 best good ADJ bracis-28410 71 13 results result NOUN bracis-28410 71 14     SPACE bracis-28410 71 15 [ [ X bracis-28410 71 16 6 6 NUM bracis-28410 71 17 , , PUNCT bracis-28410 71 18 30 30 NUM bracis-28410 71 19 , , PUNCT bracis-28410 71 20 33 33 NUM bracis-28410 71 21 , , PUNCT bracis-28410 71 22 40 40 NUM bracis-28410 71 23 , , PUNCT bracis-28410 71 24 53 53 NUM bracis-28410 71 25 , , PUNCT bracis-28410 71 26 61 61 NUM bracis-28410 71 27 , , PUNCT bracis-28410 71 28 63 63 NUM bracis-28410 71 29 , , PUNCT bracis-28410 71 30 64 64 NUM bracis-28410 71 31 ] ] PUNCT bracis-28410 71 32 and and CCONJ bracis-28410 71 33 participated participate VERB bracis-28410 71 34 in in ADP bracis-28410 71 35 several several ADJ bracis-28410 71 36 editions edition NOUN bracis-28410 71 37 of of ADP bracis-28410 71 38 the the DET bracis-28410 71 39 sisr sisr ADJ bracis-28410 71 40 competition competition NOUN bracis-28410 71 41 that that PRON bracis-28410 71 42 took take VERB bracis-28410 71 43 place place NOUN bracis-28410 71 44 during during ADP bracis-28410 71 45 the the DET bracis-28410 71 46 new new ADJ bracis-28410 71 47 trends trend NOUN bracis-28410 71 48 in in ADP bracis-28410 71 49 image image NOUN bracis-28410 71 50 restoration restoration NOUN bracis-28410 71 51 and and CCONJ bracis-28410 71 52 enhancement enhancement NOUN bracis-28410 71 53 workshop workshop NOUN bracis-28410 71 54     SPACE bracis-28410 72 1 [ [ X bracis-28410 72 2 8 8 NUM bracis-28410 72 3 , , PUNCT bracis-28410 72 4 50 50 NUM bracis-28410 72 5 , , PUNCT bracis-28410 72 6 51 51 NUM bracis-28410 72 7 , , PUNCT bracis-28410 72 8 62 62 NUM bracis-28410 72 9 ] ] PUNCT bracis-28410 72 10 . . PUNCT bracis-28410 73 1 capsules capsule NOUN bracis-28410 73 2 . . PUNCT bracis-28410 74 1 initially initially ADV bracis-28410 74 2 introduced introduce VERB bracis-28410 74 3 by by ADP bracis-28410 74 4 hinton hinton PROPN bracis-28410 74 5 et et PROPN bracis-28410 74 6 al al PROPN bracis-28410 74 7 . . PUNCT bracis-28410 74 8     SPACE bracis-28410 75 1 [ [ X bracis-28410 75 2 16 16 NUM bracis-28410 75 3 ] ] PUNCT bracis-28410 75 4 , , PUNCT bracis-28410 75 5 the the DET bracis-28410 75 6 concept concept NOUN bracis-28410 75 7 of of ADP bracis-28410 75 8 capsule capsule NOUN bracis-28410 75 9 proposes propose VERB bracis-28410 75 10 to to PART bracis-28410 75 11 solve solve VERB bracis-28410 75 12 some some PRON bracis-28410 75 13 of of ADP bracis-28410 75 14 the the DET bracis-28410 75 15 main main ADJ bracis-28410 75 16 flaws flaw NOUN bracis-28410 75 17 found find VERB bracis-28410 75 18 in in ADP bracis-28410 75 19 traditional traditional ADJ bracis-28410 75 20 convolutional convolutional ADJ bracis-28410 75 21 networks network NOUN bracis-28410 75 22 : : PUNCT bracis-28410 75 23 inability inability NOUN bracis-28410 75 24 to to PART bracis-28410 75 25 identify identify VERB bracis-28410 75 26 spatial spatial ADJ bracis-28410 75 27 hierarchy hierarchy NOUN bracis-28410 75 28 between between ADP bracis-28410 75 29 elements element NOUN bracis-28410 75 30 and and CCONJ bracis-28410 75 31 lack lack NOUN bracis-28410 75 32 of of ADP bracis-28410 75 33 rotation rotation NOUN bracis-28410 75 34 invariance invariance NOUN bracis-28410 75 35 . . PUNCT bracis-28410 76 1 hinton hinton PROPN bracis-28410 76 2 et et PROPN bracis-28410 76 3 al al PROPN bracis-28410 76 4 . . PROPN bracis-28410 77 1 conclude conclude VERB bracis-28410 77 2 that that SCONJ bracis-28410 77 3 , , PUNCT bracis-28410 77 4 after after ADP bracis-28410 77 5 several several ADJ bracis-28410 77 6 stages stage NOUN bracis-28410 77 7 of of ADP bracis-28410 77 8 subsampling subsample VERB bracis-28410 77 9 , , PUNCT bracis-28410 77 10 these these DET bracis-28410 77 11 networks network NOUN bracis-28410 77 12 lose lose VERB bracis-28410 77 13 information information NOUN bracis-28410 77 14 that that PRON bracis-28410 77 15 makes make VERB bracis-28410 77 16 them they PRON bracis-28410 77 17 possible possible ADJ bracis-28410 77 18 to to PART bracis-28410 77 19 identify identify VERB bracis-28410 77 20 the the DET bracis-28410 77 21 spatial spatial ADJ bracis-28410 77 22 relationship relationship NOUN bracis-28410 77 23 between between ADP bracis-28410 77 24 the the DET bracis-28410 77 25 elements element NOUN bracis-28410 77 26 of of ADP bracis-28410 77 27 an an DET bracis-28410 77 28 image image NOUN bracis-28410 77 29 . . PUNCT bracis-28410 78 1 the the DET bracis-28410 78 2 authors author NOUN bracis-28410 78 3 argue argue VERB bracis-28410 78 4 that that SCONJ bracis-28410 78 5 , , PUNCT bracis-28410 78 6 contrary contrary ADV bracis-28410 78 7 to to ADP bracis-28410 78 8 looking look VERB bracis-28410 78 9 for for ADP bracis-28410 78 10 a a DET bracis-28410 78 11 point point NOUN bracis-28410 78 12 of of ADP bracis-28410 78 13 view view NOUN bracis-28410 78 14 invariance invariance NOUN bracis-28410 78 15 of of ADP bracis-28410 78 16 the the DET bracis-28410 78 17 neurons neuron NOUN bracis-28410 78 18 ’ ' PUNCT bracis-28410 78 19 activities activity NOUN bracis-28410 78 20 that that PRON bracis-28410 78 21 use use VERB bracis-28410 78 22 a a DET bracis-28410 78 23 single single ADJ bracis-28410 78 24 output output NOUN bracis-28410 78 25 value value NOUN bracis-28410 78 26 , , PUNCT bracis-28410 78 27 neural neural ADJ bracis-28410 78 28 networks network NOUN bracis-28410 78 29 should should AUX bracis-28410 78 30 use use VERB bracis-28410 78 31 local local ADJ bracis-28410 78 32 “ " PUNCT bracis-28410 78 33 capsules capsule NOUN bracis-28410 78 34 ” " PUNCT bracis-28410 78 35 which which PRON bracis-28410 78 36 learn learn VERB bracis-28410 78 37 to to PART bracis-28410 78 38 recognize recognize VERB bracis-28410 78 39 a a DET bracis-28410 78 40 visual visual ADJ bracis-28410 78 41 entity entity NOUN bracis-28410 78 42 implicitly implicitly ADV bracis-28410 78 43 under under ADP bracis-28410 78 44 a a DET bracis-28410 78 45 limited limited ADJ bracis-28410 78 46 domain domain NOUN bracis-28410 78 47 of of ADP bracis-28410 78 48 viewing view VERB bracis-28410 78 49 conditions condition NOUN bracis-28410 78 50 and and CCONJ bracis-28410 78 51 deformations deformation NOUN bracis-28410 78 52 . . PUNCT bracis-28410 79 1 capsule capsule NOUN bracis-28410 79 2 structures structure NOUN bracis-28410 79 3 encode encode VERB bracis-28410 79 4 complex complex ADJ bracis-28410 79 5 calculations calculation NOUN bracis-28410 79 6 into into ADP bracis-28410 79 7 a a DET bracis-28410 79 8 small small ADJ bracis-28410 79 9 , , PUNCT bracis-28410 79 10 highly highly ADV bracis-28410 79 11 informative informative ADJ bracis-28410 79 12 output output NOUN bracis-28410 79 13 vector vector NOUN bracis-28410 79 14 . . PUNCT bracis-28410 80 1 this this DET bracis-28410 80 2 vector vector NOUN bracis-28410 80 3 contains contain VERB bracis-28410 80 4 information information NOUN bracis-28410 80 5 such such ADJ bracis-28410 80 6 as as ADP bracis-28410 80 7 the the DET bracis-28410 80 8 probability probability NOUN bracis-28410 80 9 of of ADP bracis-28410 80 10 that that DET bracis-28410 80 11 entity entity NOUN bracis-28410 80 12 being be AUX bracis-28410 80 13 present present ADJ bracis-28410 80 14 in in ADP bracis-28410 80 15 a a DET bracis-28410 80 16 compact compact ADJ bracis-28410 80 17 domain domain NOUN bracis-28410 80 18 . . PUNCT bracis-28410 81 1 in in ADP bracis-28410 81 2 addition addition NOUN bracis-28410 81 3 , , PUNCT bracis-28410 81 4 it it PRON bracis-28410 81 5 comprises comprise VERB bracis-28410 81 6 a a DET bracis-28410 81 7 set set NOUN bracis-28410 81 8 of of ADP bracis-28410 81 9 instantiation instantiation NOUN bracis-28410 81 10 parameters parameter NOUN bracis-28410 81 11 , , PUNCT bracis-28410 81 12 which which PRON bracis-28410 81 13 would would AUX bracis-28410 81 14 include include VERB bracis-28410 81 15 deformation deformation NOUN bracis-28410 81 16 , , PUNCT bracis-28410 81 17 pose pose NOUN bracis-28410 81 18 ( ( PUNCT bracis-28410 81 19 position position NOUN bracis-28410 81 20 , , PUNCT bracis-28410 81 21 size size NOUN bracis-28410 81 22 , , PUNCT bracis-28410 81 23 orientation orientation NOUN bracis-28410 81 24 ) ) PUNCT bracis-28410 81 25 , , PUNCT bracis-28410 81 26 hue hue NOUN bracis-28410 81 27 , , PUNCT bracis-28410 81 28 texture texture ADJ bracis-28410 81 29 , , PUNCT bracis-28410 81 30 and and CCONJ bracis-28410 81 31 illumination illumination NOUN bracis-28410 81 32 condition condition NOUN bracis-28410 81 33 of of ADP bracis-28410 81 34 the the DET bracis-28410 81 35 visual visual ADJ bracis-28410 81 36 entity entity NOUN bracis-28410 81 37 relative relative ADJ bracis-28410 81 38 to to ADP bracis-28410 81 39 the the DET bracis-28410 81 40 version version NOUN bracis-28410 81 41 learned learn VERB bracis-28410 81 42 by by ADP bracis-28410 81 43 the the DET bracis-28410 81 44 capsule capsule NOUN bracis-28410 81 45     SPACE bracis-28410 82 1 [ [ X bracis-28410 82 2 41 41 NUM bracis-28410 82 3 ] ] PUNCT bracis-28410 82 4 . . PUNCT bracis-28410 83 1 although although SCONJ bracis-28410 83 2 idealized idealize VERB bracis-28410 83 3 by by ADP bracis-28410 83 4 hinton hinton PROPN bracis-28410 83 5 et et PROPN bracis-28410 83 6 al al PROPN bracis-28410 83 7 . . PROPN bracis-28410 83 8 , , PUNCT bracis-28410 83 9 the the DET bracis-28410 83 10 first first ADJ bracis-28410 83 11 successful successful ADJ bracis-28410 83 12 implementation implementation NOUN bracis-28410 83 13 of of ADP bracis-28410 83 14 the the DET bracis-28410 83 15 capsule capsule NOUN bracis-28410 83 16 concept concept NOUN bracis-28410 83 17 was be AUX bracis-28410 83 18 made make VERB bracis-28410 83 19 by by ADP bracis-28410 83 20 sabour sabour PROPN bracis-28410 83 21 et et PROPN bracis-28410 83 22 al al PROPN bracis-28410 83 23 . . PUNCT bracis-28410 83 24     SPACE bracis-28410 84 1 [ [ X bracis-28410 84 2 41 41 NUM bracis-28410 84 3 ] ] PUNCT bracis-28410 84 4 . . PUNCT bracis-28410 85 1 in in ADP bracis-28410 85 2 their their PRON bracis-28410 85 3 work work NOUN bracis-28410 85 4 , , PUNCT bracis-28410 85 5 the the DET bracis-28410 85 6 authors author NOUN bracis-28410 85 7 created create VERB bracis-28410 85 8 a a DET bracis-28410 85 9 three three NUM bracis-28410 85 10 - - PUNCT bracis-28410 85 11 layer layer NOUN bracis-28410 85 12 capsule capsule NOUN bracis-28410 85 13 network network NOUN bracis-28410 85 14 that that PRON bracis-28410 85 15 achieved achieve VERB bracis-28410 85 16 comparable comparable ADJ bracis-28410 85 17 results result NOUN bracis-28410 85 18 with with ADP bracis-28410 85 19 the the DET bracis-28410 85 20 best good ADJ bracis-28410 85 21 results result NOUN bracis-28410 85 22 in in ADP bracis-28410 85 23 the the DET bracis-28410 85 24 mnist mnist NOUN bracis-28410 85 25     SPACE bracis-28410 86 1 [ [ X bracis-28410 86 2 29 29 NUM bracis-28410 86 3 ] ] PUNCT bracis-28410 86 4 digit digit NOUN bracis-28410 86 5 classification classification NOUN bracis-28410 86 6 problem problem NOUN bracis-28410 86 7 – – PUNCT bracis-28410 86 8 previously previously ADV bracis-28410 86 9 achieved achieve VERB bracis-28410 86 10 by by ADP bracis-28410 86 11 deeper deep ADJ bracis-28410 86 12 networks network NOUN bracis-28410 86 13 only only ADV bracis-28410 86 14 . . PUNCT bracis-28410 87 1 for for ADP bracis-28410 87 2 this this PRON bracis-28410 87 3 , , PUNCT bracis-28410 87 4 sabour sabour PROPN bracis-28410 87 5 et et PROPN bracis-28410 87 6 al al PROPN bracis-28410 87 7 . . PUNCT bracis-28410 87 8     SPACE bracis-28410 88 1 [ [ X bracis-28410 88 2 41 41 NUM bracis-28410 88 3 ] ] PUNCT bracis-28410 88 4 developed develop VERB bracis-28410 88 5 two two NUM bracis-28410 88 6 innovative innovative ADJ bracis-28410 88 7 concepts concept NOUN bracis-28410 88 8 : : PUNCT bracis-28410 88 9 dynamic dynamic ADJ bracis-28410 88 10 routing routing NOUN bracis-28410 88 11 and and CCONJ bracis-28410 88 12 a a DET bracis-28410 88 13 new new ADJ bracis-28410 88 14 activation activation NOUN bracis-28410 88 15 function function NOUN bracis-28410 88 16 . . PUNCT bracis-28410 89 1 leveraged leverage VERB bracis-28410 89 2 by by ADP bracis-28410 89 3 sabour sabour PROPN bracis-28410 89 4 et et PROPN bracis-28410 89 5 al al PROPN bracis-28410 89 6 . . PUNCT bracis-28410 89 7     SPACE bracis-28410 90 1 [ [ X bracis-28410 90 2 41 41 NUM bracis-28410 90 3 ] ] PUNCT bracis-28410 90 4 , , PUNCT bracis-28410 90 5 many many ADJ bracis-28410 90 6 other other ADJ bracis-28410 90 7 authors author NOUN bracis-28410 90 8 have have AUX bracis-28410 90 9 enhanced enhance VERB bracis-28410 90 10 the the DET bracis-28410 90 11 concept concept NOUN bracis-28410 90 12 of of ADP bracis-28410 90 13 capsules capsule NOUN bracis-28410 90 14 . . PUNCT bracis-28410 91 1 hinton hinton PROPN bracis-28410 91 2 et et PROPN bracis-28410 91 3 al al PROPN bracis-28410 91 4 . . PUNCT bracis-28410 91 5     SPACE bracis-28410 92 1 [ [ X bracis-28410 92 2 15 15 NUM bracis-28410 92 3 ] ] PUNCT bracis-28410 92 4 proposed propose VERB bracis-28410 92 5 a a DET bracis-28410 92 6 new new ADJ bracis-28410 92 7 type type NOUN bracis-28410 92 8 of of ADP bracis-28410 92 9 capsule capsule NOUN bracis-28410 92 10 composed compose VERB bracis-28410 92 11 of of ADP bracis-28410 92 12 a a DET bracis-28410 92 13 logistic logistic ADJ bracis-28410 92 14 unit unit NOUN bracis-28410 92 15 that that PRON bracis-28410 92 16 indicates indicate VERB bracis-28410 92 17 the the DET bracis-28410 92 18 probability probability NOUN bracis-28410 92 19 of of ADP bracis-28410 92 20 the the DET bracis-28410 92 21 presence presence NOUN bracis-28410 92 22 of of ADP bracis-28410 92 23 an an DET bracis-28410 92 24 entity entity NOUN bracis-28410 92 25 and and CCONJ bracis-28410 92 26 a a DET bracis-28410 92 27 pose pose NOUN bracis-28410 92 28 matrix matrix NOUN bracis-28410 92 29 of of ADP bracis-28410 92 30 \(4 \(4 PRON bracis-28410 92 31 \times \time NOUN bracis-28410 92 32 4\ 4\ NUM bracis-28410 92 33 ) ) PUNCT bracis-28410 92 34 representing represent VERB bracis-28410 92 35 the the DET bracis-28410 92 36 pose pose NOUN bracis-28410 92 37 of of ADP bracis-28410 92 38 that that DET bracis-28410 92 39 entity entity NOUN bracis-28410 92 40 . . PUNCT bracis-28410 93 1 the the DET bracis-28410 93 2 authors author NOUN bracis-28410 93 3 also also ADV bracis-28410 93 4 introduced introduce VERB bracis-28410 93 5 a a DET bracis-28410 93 6 new new ADJ bracis-28410 93 7 routing routing NOUN bracis-28410 93 8 algorithm algorithm NOUN bracis-28410 93 9 , , PUNCT bracis-28410 93 10 which which PRON bracis-28410 93 11 allows allow VERB bracis-28410 93 12 the the DET bracis-28410 93 13 outputs output NOUN bracis-28410 93 14 of of ADP bracis-28410 93 15 the the DET bracis-28410 93 16 capsules capsule NOUN bracis-28410 93 17 to to PART bracis-28410 93 18 be be AUX bracis-28410 93 19 routed route VERB bracis-28410 93 20 to to ADP bracis-28410 93 21 those those PRON bracis-28410 93 22 of of ADP bracis-28410 93 23 the the DET bracis-28410 93 24 next next ADJ bracis-28410 93 25 layer layer NOUN bracis-28410 93 26 so so SCONJ bracis-28410 93 27 that that SCONJ bracis-28410 93 28 the the DET bracis-28410 93 29 active active ADJ bracis-28410 93 30 capsules capsule NOUN bracis-28410 93 31 receive receive VERB bracis-28410 93 32 a a DET bracis-28410 93 33 group group NOUN bracis-28410 93 34 of of ADP bracis-28410 93 35 votes vote NOUN bracis-28410 93 36 from from ADP bracis-28410 93 37 similar similar ADJ bracis-28410 93 38 poses pose NOUN bracis-28410 93 39 . . PUNCT bracis-28410 94 1 hinton hinton PROPN bracis-28410 94 2 et et PROPN bracis-28410 94 3 al al PROPN bracis-28410 94 4 . . PROPN bracis-28410 94 5 showed show VERB bracis-28410 94 6 that that SCONJ bracis-28410 94 7 their their PRON bracis-28410 94 8 model model NOUN bracis-28410 94 9 surpasses surpass VERB bracis-28410 94 10 the the DET bracis-28410 94 11 best good ADJ bracis-28410 94 12 result result NOUN bracis-28410 94 13 in in ADP bracis-28410 94 14 the the DET bracis-28410 94 15 smallnorb smallnorb NOUN bracis-28410 94 16 dataset dataset NOUN bracis-28410 94 17 , , PUNCT bracis-28410 94 18 reducing reduce VERB bracis-28410 94 19 the the DET bracis-28410 94 20 number number NOUN bracis-28410 94 21 of of ADP bracis-28410 94 22 errors error NOUN bracis-28410 94 23 by by ADP bracis-28410 94 24 more more ADJ bracis-28410 94 25 than than ADP bracis-28410 94 26 40 40 NUM bracis-28410 94 27 % % NOUN bracis-28410 94 28 while while SCONJ bracis-28410 94 29 being be AUX bracis-28410 94 30 significantly significantly ADV bracis-28410 94 31 more more ADV bracis-28410 94 32 resistant resistant ADJ bracis-28410 94 33 to to ADP bracis-28410 94 34 white white ADJ bracis-28410 94 35 - - PUNCT bracis-28410 94 36 box box NOUN bracis-28410 94 37 adversarial adversarial ADJ bracis-28410 94 38 attacks attack NOUN bracis-28410 94 39 . . PUNCT bracis-28410 95 1 a a DET bracis-28410 95 2 remarkable remarkable ADJ bracis-28410 95 3 work work NOUN bracis-28410 95 4 , , PUNCT bracis-28410 95 5 which which PRON bracis-28410 95 6 made make VERB bracis-28410 95 7 possible possible ADJ bracis-28410 95 8 the the DET bracis-28410 95 9 development development NOUN bracis-28410 95 10 of of ADP bracis-28410 95 11 our our PRON bracis-28410 95 12 solution solution NOUN bracis-28410 95 13 , , PUNCT bracis-28410 95 14 was be AUX bracis-28410 95 15 developed develop VERB bracis-28410 95 16 by by ADP bracis-28410 95 17 lalonde lalonde NOUN bracis-28410 95 18 and and CCONJ bracis-28410 95 19 bagci bagci NOUN bracis-28410 95 20     SPACE bracis-28410 96 1 [ [ X bracis-28410 96 2 28 28 NUM bracis-28410 96 3 ] ] PUNCT bracis-28410 96 4 . . PUNCT bracis-28410 97 1 the the DET bracis-28410 97 2 authors author NOUN bracis-28410 97 3 expanded expand VERB bracis-28410 97 4 the the DET bracis-28410 97 5 use use NOUN bracis-28410 97 6 of of ADP bracis-28410 97 7 capsules capsule NOUN bracis-28410 97 8 for for ADP bracis-28410 97 9 the the DET bracis-28410 97 10 problem problem NOUN bracis-28410 97 11 of of ADP bracis-28410 97 12 object object NOUN bracis-28410 97 13 segmentation segmentation NOUN bracis-28410 97 14 and and CCONJ bracis-28410 97 15 made make VERB bracis-28410 97 16 innovations innovation NOUN bracis-28410 97 17 that that PRON bracis-28410 97 18 allowed allow VERB bracis-28410 97 19 , , PUNCT bracis-28410 97 20 among among ADP bracis-28410 97 21 other other ADJ bracis-28410 97 22 gains gain NOUN bracis-28410 97 23 , , PUNCT bracis-28410 97 24 to to PART bracis-28410 97 25 increase increase VERB bracis-28410 97 26 the the DET bracis-28410 97 27 data datum NOUN bracis-28410 97 28 processing processing NOUN bracis-28410 97 29 capacity capacity NOUN bracis-28410 97 30 of of ADP bracis-28410 97 31 the the DET bracis-28410 97 32 capsule capsule NOUN bracis-28410 97 33 network network NOUN bracis-28410 97 34 , , PUNCT bracis-28410 97 35 increasing increase VERB bracis-28410 97 36 inputs input NOUN bracis-28410 97 37 from from ADP bracis-28410 97 38 \(32 \(32 NOUN bracis-28410 97 39 \times \times ADP bracis-28410 97 40 32\ 32\ NUM bracis-28410 97 41 ) ) PUNCT bracis-28410 97 42 to to ADP bracis-28410 97 43 \(512 \(512 PROPN bracis-28410 97 44 \times \times ADP bracis-28410 97 45 512\ 512\ NUM bracis-28410 97 46 ) ) PUNCT bracis-28410 97 47 pixels pixel NOUN bracis-28410 97 48 . . PUNCT bracis-28410 98 1 most most ADV bracis-28410 98 2 significantly significantly ADV bracis-28410 98 3 , , PUNCT bracis-28410 98 4 they they PRON bracis-28410 98 5 advanced advance VERB bracis-28410 98 6 the the DET bracis-28410 98 7 state state NOUN bracis-28410 98 8 - - PUNCT bracis-28410 98 9 of of ADP bracis-28410 98 10 - - PUNCT bracis-28410 98 11 the the DET bracis-28410 98 12 - - PUNCT bracis-28410 98 13 art art NOUN bracis-28410 98 14 in in ADP bracis-28410 98 15 the the DET bracis-28410 98 16 problem problem NOUN bracis-28410 98 17 of of ADP bracis-28410 98 18 segmentation segmentation NOUN bracis-28410 98 19 of of ADP bracis-28410 98 20 lung lung NOUN bracis-28410 98 21 pathologies pathology NOUN bracis-28410 98 22 from from ADP bracis-28410 98 23 computed computed ADJ bracis-28410 98 24 tomography tomography NOUN bracis-28410 98 25 , , PUNCT bracis-28410 98 26 while while SCONJ bracis-28410 98 27 reducing reduce VERB bracis-28410 98 28 the the DET bracis-28410 98 29 number number NOUN bracis-28410 98 30 of of ADP bracis-28410 98 31 parameters parameter NOUN bracis-28410 98 32 by by ADP bracis-28410 98 33 approximately approximately ADV bracis-28410 98 34 38 38 NUM bracis-28410 98 35 % % NOUN bracis-28410 98 36 . . PUNCT bracis-28410 99 1 in in ADP bracis-28410 99 2 particular particular ADJ bracis-28410 99 3 , , PUNCT bracis-28410 99 4 the the DET bracis-28410 99 5 authors author NOUN bracis-28410 99 6 modified modify VERB bracis-28410 99 7 the the DET bracis-28410 99 8 capsule capsule NOUN bracis-28410 99 9 routing route VERB bracis-28410 99 10 algorithm algorithm NOUN bracis-28410 99 11 and and CCONJ bracis-28410 99 12 the the DET bracis-28410 99 13 reconstruction reconstruction NOUN bracis-28410 99 14 part part NOUN bracis-28410 99 15 , , PUNCT bracis-28410 99 16 and and CCONJ bracis-28410 99 17 modified modify VERB bracis-28410 99 18 the the DET bracis-28410 99 19 concept concept NOUN bracis-28410 99 20 of of ADP bracis-28410 99 21 convolutional convolutional ADJ bracis-28410 99 22 capsules capsule NOUN bracis-28410 99 23 . . PUNCT bracis-28410 100 1 recently recently ADV bracis-28410 100 2 , , PUNCT bracis-28410 100 3 a a DET bracis-28410 100 4 few few ADJ bracis-28410 100 5 authors author NOUN bracis-28410 100 6 have have AUX bracis-28410 100 7 employed employ VERB bracis-28410 100 8 capsules capsule NOUN bracis-28410 100 9 in in ADP bracis-28410 100 10 their their PRON bracis-28410 100 11 solutions solution NOUN bracis-28410 100 12 for for ADP bracis-28410 100 13 problems problem NOUN bracis-28410 100 14 involving involve VERB bracis-28410 100 15 lr lr NOUN bracis-28410 100 16 images image NOUN bracis-28410 100 17     SPACE bracis-28410 100 18 [ [ X bracis-28410 100 19 18 18 NUM bracis-28410 100 20 , , PUNCT bracis-28410 100 21 35 35 NUM bracis-28410 100 22 , , PUNCT bracis-28410 100 23 46 46 NUM bracis-28410 100 24 ] ] PUNCT bracis-28410 100 25 . . PUNCT bracis-28410 101 1 it it PRON bracis-28410 101 2 is be AUX bracis-28410 101 3 worth worth ADJ bracis-28410 101 4 noting note VERB bracis-28410 101 5 that that SCONJ bracis-28410 101 6 most most ADJ bracis-28410 101 7 of of ADP bracis-28410 101 8 these these DET bracis-28410 101 9 solutions solution NOUN bracis-28410 101 10 only only ADV bracis-28410 101 11 made make VERB bracis-28410 101 12 small small ADJ bracis-28410 101 13 changes change NOUN bracis-28410 101 14 to to ADP bracis-28410 101 15 the the DET bracis-28410 101 16 first first ADJ bracis-28410 101 17 capsule capsule NOUN bracis-28410 101 18 networks network NOUN bracis-28410 101 19 introduced introduce VERB bracis-28410 101 20 by by ADP bracis-28410 101 21 sabour sabour PROPN bracis-28410 101 22 et et PROPN bracis-28410 101 23 al al PROPN bracis-28410 101 24 . . PUNCT bracis-28410 101 25     SPACE bracis-28410 102 1 [ [ X bracis-28410 102 2 41 41 NUM bracis-28410 102 3 ] ] PUNCT bracis-28410 102 4 and and CCONJ bracis-28410 102 5 hinton hinton PROPN bracis-28410 102 6 et et PROPN bracis-28410 102 7 al al PROPN bracis-28410 102 8 . . PUNCT bracis-28410 102 9     SPACE bracis-28410 103 1 [ [ X bracis-28410 103 2 15 15 NUM bracis-28410 103 3 ] ] PUNCT bracis-28410 103 4 . . PUNCT bracis-28410 104 1 for for ADP bracis-28410 104 2 example example NOUN bracis-28410 104 3 , , PUNCT bracis-28410 104 4 majdabadi majdabadi NOUN bracis-28410 104 5 and and CCONJ bracis-28410 104 6 ko ko PROPN bracis-28410 104 7     SPACE bracis-28410 105 1 [ [ X bracis-28410 105 2 35 35 NUM bracis-28410 105 3 ] ] PUNCT bracis-28410 105 4 employed employ VERB bracis-28410 105 5 a a DET bracis-28410 105 6 two two NUM bracis-28410 105 7 - - PUNCT bracis-28410 105 8 layered layer VERB bracis-28410 105 9 capsule capsule NOUN bracis-28410 105 10 network network NOUN bracis-28410 105 11 with with ADP bracis-28410 105 12 dynamic dynamic ADJ bracis-28410 105 13 routing routing NOUN bracis-28410 105 14 as as ADP bracis-28410 105 15 the the DET bracis-28410 105 16 discriminator discriminator NOUN bracis-28410 105 17 for for ADP bracis-28410 105 18 its its PRON bracis-28410 105 19 multi multi ADJ bracis-28410 105 20 - - ADJ bracis-28410 105 21 scale scale ADJ bracis-28410 105 22 gradient gradient NOUN bracis-28410 105 23 capsule capsule NOUN bracis-28410 105 24 gan gan PROPN bracis-28410 105 25 . . PUNCT bracis-28410 106 1 leveraged leverage VERB bracis-28410 106 2 by by ADP bracis-28410 106 3 the the DET bracis-28410 106 4 matrix matrix NOUN bracis-28410 106 5 capsules capsule NOUN bracis-28410 106 6 by by ADP bracis-28410 106 7 hinton hinton PROPN bracis-28410 106 8 et et PROPN bracis-28410 106 9 al al PROPN bracis-28410 106 10 . . PUNCT bracis-28410 106 11     SPACE bracis-28410 107 1 [ [ X bracis-28410 107 2 15 15 NUM bracis-28410 107 3 ] ] PUNCT bracis-28410 107 4 , , PUNCT bracis-28410 107 5 hsu hsu PROPN bracis-28410 107 6 et et PROPN bracis-28410 107 7 al al PROPN bracis-28410 107 8 . . PUNCT bracis-28410 107 9     SPACE bracis-28410 108 1 [ [ X bracis-28410 108 2 18 18 NUM bracis-28410 108 3 ] ] PUNCT bracis-28410 108 4 created create VERB bracis-28410 108 5 two two NUM bracis-28410 108 6 different different ADJ bracis-28410 108 7 approaches approach NOUN bracis-28410 108 8 : : PUNCT bracis-28410 108 9 ( ( PUNCT bracis-28410 108 10 i i NOUN bracis-28410 108 11 ) ) PUNCT bracis-28410 108 12 capsules capsule NOUN bracis-28410 108 13 as as ADP bracis-28410 108 14 its its PRON bracis-28410 108 15 main main ADJ bracis-28410 108 16 component component NOUN bracis-28410 108 17 in in ADP bracis-28410 108 18 the the DET bracis-28410 108 19 network network NOUN bracis-28410 108 20 and and CCONJ bracis-28410 108 21 reconstructing reconstruct VERB bracis-28410 108 22 hr hr NOUN bracis-28410 108 23 images image NOUN bracis-28410 108 24 directly directly ADV bracis-28410 108 25 from from ADP bracis-28410 108 26 it it PRON bracis-28410 108 27 ( ( PUNCT bracis-28410 108 28 cirnn cirnn NOUN bracis-28410 108 29 ) ) PUNCT bracis-28410 108 30 and and CCONJ bracis-28410 108 31 ( ( PUNCT bracis-28410 108 32 ii ii NOUN bracis-28410 108 33 ) ) PUNCT bracis-28410 108 34 capsules capsule NOUN bracis-28410 108 35 for for ADP bracis-28410 108 36 the the DET bracis-28410 108 37 channel channel NOUN bracis-28410 108 38 attention attention NOUN bracis-28410 108 39 mechanism mechanism NOUN bracis-28410 108 40 ( ( PUNCT bracis-28410 108 41 carnn carnn PROPN bracis-28410 108 42 ) ) PUNCT bracis-28410 108 43 . . PUNCT bracis-28410 109 1 3 3 NUM bracis-28410 109 2 proposed propose VERB bracis-28410 109 3 method method NOUN bracis-28410 109 4 the the DET bracis-28410 109 5 proposed propose VERB bracis-28410 109 6 model model NOUN bracis-28410 109 7 , , PUNCT bracis-28410 109 8 named name VERB bracis-28410 109 9 super super ADJ bracis-28410 109 10 - - ADJ bracis-28410 109 11 resolution resolution ADJ bracis-28410 109 12 capsules capsule NOUN bracis-28410 109 13 ( ( PUNCT bracis-28410 109 14 srcaps srcap NOUN bracis-28410 109 15 ) ) PUNCT bracis-28410 109 16 , , PUNCT bracis-28410 109 17 is be AUX bracis-28410 109 18 shown show VERB bracis-28410 109 19 in in ADP bracis-28410 109 20 fig fig NOUN bracis-28410 109 21 . . PUNCT bracis-28410 109 22     SPACE bracis-28410 110 1 1 1 X bracis-28410 110 2 . . PUNCT bracis-28410 111 1 it it PRON bracis-28410 111 2 consists consist VERB bracis-28410 111 3 of of ADP bracis-28410 111 4 four four NUM bracis-28410 111 5 main main ADJ bracis-28410 111 6 parts part NOUN bracis-28410 111 7 : : PUNCT bracis-28410 111 8 an an DET bracis-28410 111 9 initial initial ADJ bracis-28410 111 10 convolutional convolutional ADJ bracis-28410 111 11 layer layer NOUN bracis-28410 111 12 , , PUNCT bracis-28410 111 13 followed follow VERB bracis-28410 111 14 by by ADP bracis-28410 111 15 b b PROPN bracis-28410 111 16 sequentially sequentially ADV bracis-28410 111 17 connected connect VERB bracis-28410 111 18 residual residual ADJ bracis-28410 111 19 dense dense ADJ bracis-28410 111 20 capsule capsule NOUN bracis-28410 111 21 blocks block NOUN bracis-28410 111 22 , , PUNCT bracis-28410 111 23 a a DET bracis-28410 111 24 new new ADJ bracis-28410 111 25 convolutional convolutional ADJ bracis-28410 111 26 layer layer NOUN bracis-28410 111 27 and and CCONJ bracis-28410 111 28 , , PUNCT bracis-28410 111 29 finally finally ADV bracis-28410 111 30 , , PUNCT bracis-28410 111 31 a a DET bracis-28410 111 32 neural neural ADJ bracis-28410 111 33 network network NOUN bracis-28410 111 34 to to PART bracis-28410 111 35 increase increase VERB bracis-28410 111 36 resolution resolution NOUN bracis-28410 111 37 . . PUNCT bracis-28410 112 1 all all DET bracis-28410 112 2 the the DET bracis-28410 112 3 convolution convolution NOUN bracis-28410 112 4 - - PUNCT bracis-28410 112 5 based base VERB bracis-28410 112 6 layers layer NOUN bracis-28410 112 7 use use VERB bracis-28410 112 8 the the DET bracis-28410 112 9 weight weight NOUN bracis-28410 112 10 normalization normalization NOUN bracis-28410 112 11 technique technique NOUN bracis-28410 112 12     SPACE bracis-28410 113 1 [ [ X bracis-28410 113 2 43 43 NUM bracis-28410 113 3 ] ] X bracis-28410 113 4 , , PUNCT bracis-28410 113 5 as as SCONJ bracis-28410 113 6 it it PRON bracis-28410 113 7 accelerates accelerate VERB bracis-28410 113 8 the the DET bracis-28410 113 9 training training NOUN bracis-28410 113 10 convergence convergence NOUN bracis-28410 113 11 and and CCONJ bracis-28410 113 12 has have VERB bracis-28410 113 13 a a DET bracis-28410 113 14 lower low ADJ bracis-28410 113 15 computational computational ADJ bracis-28410 113 16 cost cost NOUN bracis-28410 113 17 if if SCONJ bracis-28410 113 18 compared compare VERB bracis-28410 113 19 to to PART bracis-28410 113 20 batch batch VERB bracis-28410 113 21 normalization normalization NOUN bracis-28410 113 22 , , PUNCT bracis-28410 113 23 without without ADP bracis-28410 113 24 introducing introduce VERB bracis-28410 113 25 dependencies dependency NOUN bracis-28410 113 26 between between ADP bracis-28410 113 27 the the DET bracis-28410 113 28 examples example NOUN bracis-28410 113 29 of of ADP bracis-28410 113 30 the the DET bracis-28410 113 31 batch batch NOUN bracis-28410 113 32     SPACE bracis-28410 114 1 [ [ X bracis-28410 114 2 61 61 NUM bracis-28410 114 3 ] ] PUNCT bracis-28410 114 4 . . PUNCT bracis-28410 115 1 the the DET bracis-28410 115 2 first first ADJ bracis-28410 115 3 convolutional convolutional ADJ bracis-28410 115 4 layer layer NOUN bracis-28410 115 5 , , PUNCT bracis-28410 115 6 conv conv ADJ bracis-28410 115 7 act act NOUN bracis-28410 115 8 in in ADP bracis-28410 115 9 fig fig NOUN bracis-28410 115 10 . . PUNCT bracis-28410 115 11     SPACE bracis-28410 115 12 1 1 NUM bracis-28410 115 13 , , PUNCT bracis-28410 115 14 generates generate VERB bracis-28410 115 15 f f PROPN bracis-28410 115 16 filters filter NOUN bracis-28410 115 17 from from ADP bracis-28410 115 18 convolutional convolutional ADJ bracis-28410 115 19 kernels kernel NOUN bracis-28410 115 20 of of ADP bracis-28410 115 21 size size NOUN bracis-28410 115 22 \(k \(k NOUN bracis-28410 115 23 \times \times ADP bracis-28410 115 24 k\ k\ PROPN bracis-28410 115 25 ) ) PUNCT bracis-28410 115 26 with with ADP bracis-28410 115 27 stride stride NOUN bracis-28410 115 28 st st NOUN bracis-28410 115 29 and and CCONJ bracis-28410 115 30 padding padding NOUN bracis-28410 115 31 p p NOUN bracis-28410 115 32 , , PUNCT bracis-28410 115 33 followed follow VERB bracis-28410 115 34 by by ADP bracis-28410 115 35 an an DET bracis-28410 115 36 activation activation NOUN bracis-28410 115 37 function function NOUN bracis-28410 115 38 act act PROPN bracis-28410 115 39 . . PUNCT bracis-28410 116 1 this this DET bracis-28410 116 2 layer layer NOUN bracis-28410 116 3 is be AUX bracis-28410 116 4 responsible responsible ADJ bracis-28410 116 5 for for ADP bracis-28410 116 6 converting convert VERB bracis-28410 116 7 pixel pixel ADJ bracis-28410 116 8 intensities intensity NOUN bracis-28410 116 9 to to ADP bracis-28410 116 10 local local ADJ bracis-28410 116 11 resource resource NOUN bracis-28410 116 12 detector detector NOUN bracis-28410 116 13 activations activation NOUN bracis-28410 116 14 that that PRON bracis-28410 116 15 are be AUX bracis-28410 116 16 used use VERB bracis-28410 116 17 as as ADP bracis-28410 116 18 inputs input NOUN bracis-28410 116 19 to to ADP bracis-28410 116 20 the the DET bracis-28410 116 21 next next ADJ bracis-28410 116 22 step step NOUN bracis-28410 116 23 in in ADP bracis-28410 116 24 the the DET bracis-28410 116 25 capsule capsule NOUN bracis-28410 116 26 blocks block NOUN bracis-28410 116 27 . . PUNCT bracis-28410 117 1 fig fig NOUN bracis-28410 117 2 . . PUNCT bracis-28410 118 1 1 1 X bracis-28410 118 2 . . X bracis-28410 118 3 diagram diagram NOUN bracis-28410 118 4 of of ADP bracis-28410 118 5 srcaps srcap NOUN bracis-28410 118 6 model model NOUN bracis-28410 118 7 . . PUNCT bracis-28410 119 1 full full ADJ bracis-28410 119 2 size size NOUN bracis-28410 119 3 image image NOUN bracis-28410 119 4 the the DET bracis-28410 119 5 residual residual ADJ bracis-28410 119 6 dense dense ADJ bracis-28410 119 7 capsule capsule NOUN bracis-28410 119 8 blocks block NOUN bracis-28410 119 9 , , PUNCT bracis-28410 119 10 rdcbs rdcbs PROPN bracis-28410 119 11 , , PUNCT bracis-28410 119 12 are be AUX bracis-28410 119 13 composed compose VERB bracis-28410 119 14 of of ADP bracis-28410 119 15 l l NOUN bracis-28410 119 16 convolutional convolutional ADJ bracis-28410 119 17 capsule capsule NOUN bracis-28410 119 18 layers layer NOUN bracis-28410 119 19 followed follow VERB bracis-28410 119 20 by by ADP bracis-28410 119 21 an an DET bracis-28410 119 22 activation activation NOUN bracis-28410 119 23 function function NOUN bracis-28410 119 24 act act NOUN bracis-28410 119 25 , , PUNCT bracis-28410 119 26 with with ADP bracis-28410 119 27 residual residual ADJ bracis-28410 119 28 connection connection NOUN bracis-28410 119 29 to to ADP bracis-28410 119 30 their their PRON bracis-28410 119 31 inputs input NOUN bracis-28410 119 32 , , PUNCT bracis-28410 119 33 sequentially sequentially ADV bracis-28410 119 34 connected connect VERB bracis-28410 119 35 . . PUNCT bracis-28410 120 1 the the DET bracis-28410 120 2 outputs output NOUN bracis-28410 120 3 of of ADP bracis-28410 120 4 these these DET bracis-28410 120 5 layers layer NOUN bracis-28410 120 6 are be AUX bracis-28410 120 7 concatenated concatenate VERB bracis-28410 120 8 , , PUNCT bracis-28410 120 9 forming form VERB bracis-28410 120 10 a a DET bracis-28410 120 11 dense dense ADJ bracis-28410 120 12 connection connection NOUN bracis-28410 120 13 , , PUNCT bracis-28410 120 14 followed follow VERB bracis-28410 120 15 by by ADP bracis-28410 120 16 a a DET bracis-28410 120 17 convolutional convolutional ADJ bracis-28410 120 18 layer layer NOUN bracis-28410 120 19 , , PUNCT bracis-28410 120 20 as as SCONJ bracis-28410 120 21 shown show VERB bracis-28410 120 22 in in ADP bracis-28410 120 23 fig fig NOUN bracis-28410 120 24 . . PUNCT bracis-28410 120 25     SPACE bracis-28410 121 1 2 2 X bracis-28410 121 2 . . PUNCT bracis-28410 122 1 this this DET bracis-28410 122 2 convolutional convolutional ADJ bracis-28410 122 3 layer layer NOUN bracis-28410 122 4 , , PUNCT bracis-28410 122 5 with with ADP bracis-28410 122 6 kernels kernel NOUN bracis-28410 122 7 \(1 \(1 PROPN bracis-28410 122 8 \times \times ADP bracis-28410 122 9 1\ 1\ NUM bracis-28410 122 10 ) ) PUNCT bracis-28410 122 11 , , PUNCT bracis-28410 122 12 stride stride NOUN bracis-28410 122 13 1 1 NUM bracis-28410 122 14 and and CCONJ bracis-28410 122 15 padding padding NOUN bracis-28410 122 16 0 0 NUM bracis-28410 122 17 , , PUNCT bracis-28410 122 18 acts act VERB bracis-28410 122 19 as as ADP bracis-28410 122 20 a a DET bracis-28410 122 21 weighted weighted ADJ bracis-28410 122 22 sum sum NOUN bracis-28410 122 23 between between ADP bracis-28410 122 24 the the DET bracis-28410 122 25 various various ADJ bracis-28410 122 26 filters filter NOUN bracis-28410 122 27 , , PUNCT bracis-28410 122 28 allowing allow VERB bracis-28410 122 29 the the DET bracis-28410 122 30 network network NOUN bracis-28410 122 31 to to PART bracis-28410 122 32 learn learn VERB bracis-28410 122 33 which which DET bracis-28410 122 34 filters filter NOUN bracis-28410 122 35 are be AUX bracis-28410 122 36 more more ADV bracis-28410 122 37 important important ADJ bracis-28410 122 38 , , PUNCT bracis-28410 122 39 thus thus ADV bracis-28410 122 40 reducing reduce VERB bracis-28410 122 41 dimensionality dimensionality NOUN bracis-28410 122 42 more more ADV bracis-28410 122 43 efficiently efficiently ADV bracis-28410 122 44     SPACE bracis-28410 123 1 [ [ X bracis-28410 123 2 34 34 NUM bracis-28410 123 3 , , PUNCT bracis-28410 123 4 45 45 NUM bracis-28410 123 5 , , PUNCT bracis-28410 123 6 49 49 NUM bracis-28410 123 7 ] ] PUNCT bracis-28410 123 8 . . PUNCT bracis-28410 124 1 the the DET bracis-28410 124 2 output output NOUN bracis-28410 124 3 of of ADP bracis-28410 124 4 the the DET bracis-28410 124 5 rdcb rdcb NOUN bracis-28410 124 6 is be AUX bracis-28410 124 7 weighted weight VERB bracis-28410 124 8 by by ADP bracis-28410 124 9 a a DET bracis-28410 124 10 residual residual ADJ bracis-28410 124 11 scale scale NOUN bracis-28410 124 12 constant constant ADJ bracis-28410 124 13 . . PUNCT bracis-28410 125 1 fig fig NOUN bracis-28410 125 2 . . PUNCT bracis-28410 126 1 2 2 X bracis-28410 126 2 . . X bracis-28410 126 3 rdcb rdcb ADJ bracis-28410 126 4 diagram diagram PROPN bracis-28410 126 5 . . PUNCT bracis-28410 127 1 full full ADJ bracis-28410 127 2 size size NOUN bracis-28410 127 3 image image NOUN bracis-28410 127 4 all all PRON bracis-28410 127 5 capsules capsule VERB bracis-28410 127 6 layers layer NOUN bracis-28410 127 7 within within ADP bracis-28410 127 8 an an DET bracis-28410 127 9 rdcb rdcb ADJ bracis-28410 127 10 module module NOUN bracis-28410 127 11 have have VERB bracis-28410 127 12 the the DET bracis-28410 127 13 same same ADJ bracis-28410 127 14 parameters parameter NOUN bracis-28410 127 15 : : PUNCT bracis-28410 127 16 the the DET bracis-28410 127 17 number number NOUN bracis-28410 127 18 of of ADP bracis-28410 127 19 capsules capsule NOUN bracis-28410 127 20 per per ADP bracis-28410 127 21 layer layer NOUN bracis-28410 127 22 c c NOUN bracis-28410 127 23 , , PUNCT bracis-28410 127 24 amount amount NOUN bracis-28410 127 25 of of ADP bracis-28410 127 26 filters filter NOUN bracis-28410 127 27 f f PROPN bracis-28410 127 28 , , PUNCT bracis-28410 127 29 kernel kernel PROPN bracis-28410 127 30 size size NOUN bracis-28410 127 31 k k PROPN bracis-28410 127 32 , , PUNCT bracis-28410 127 33 stride stride ADJ bracis-28410 127 34 st st NOUN bracis-28410 127 35 and and CCONJ bracis-28410 127 36 padding padding NOUN bracis-28410 127 37 p. p. NOUN bracis-28410 127 38 our our PRON bracis-28410 127 39 capsules capsule NOUN bracis-28410 127 40 employ employ VERB bracis-28410 127 41 the the DET bracis-28410 127 42 routing routing NOUN bracis-28410 127 43 algorithm algorithm NOUN bracis-28410 127 44 suggested suggest VERB bracis-28410 127 45 by by ADP bracis-28410 127 46 lalonde lalonde NOUN bracis-28410 127 47 and and CCONJ bracis-28410 127 48 bagci bagci NOUN bracis-28410 127 49     SPACE bracis-28410 128 1 [ [ X bracis-28410 128 2 28 28 NUM bracis-28410 128 3 ] ] PUNCT bracis-28410 128 4 , , PUNCT bracis-28410 128 5 because because SCONJ bracis-28410 128 6 it it PRON bracis-28410 128 7 provides provide VERB bracis-28410 128 8 an an DET bracis-28410 128 9 efficient efficient ADJ bracis-28410 128 10 version version NOUN bracis-28410 128 11 of of ADP bracis-28410 128 12 the the DET bracis-28410 128 13 original original ADJ bracis-28410 128 14 capsule capsule NOUN bracis-28410 128 15 definition definition NOUN bracis-28410 128 16 , , PUNCT bracis-28410 128 17 as as SCONJ bracis-28410 128 18 we we PRON bracis-28410 128 19 mentioned mention VERB bracis-28410 128 20 before before ADV bracis-28410 128 21 . . PUNCT bracis-28410 129 1 this this DET bracis-28410 129 2 algorithm algorithm NOUN bracis-28410 129 3 differs differ VERB bracis-28410 129 4 from from ADP bracis-28410 129 5 the the DET bracis-28410 129 6 routing routing NOUN bracis-28410 129 7 - - PUNCT bracis-28410 129 8 by by ADP bracis-28410 129 9 - - PUNCT bracis-28410 129 10 agreement agreement NOUN bracis-28410 129 11 implementation implementation NOUN bracis-28410 129 12 by by ADP bracis-28410 129 13 sabour sabour PROPN bracis-28410 129 14 et et PROPN bracis-28410 129 15 al al PROPN bracis-28410 129 16 . . PUNCT bracis-28410 129 17     SPACE bracis-28410 130 1 [ [ X bracis-28410 130 2 41 41 NUM bracis-28410 130 3 ] ] PUNCT bracis-28410 130 4 in in ADP bracis-28410 130 5 two two NUM bracis-28410 130 6 ways way NOUN bracis-28410 130 7 . . PUNCT bracis-28410 131 1 first first ADV bracis-28410 131 2 , , PUNCT bracis-28410 131 3 we we PRON bracis-28410 131 4 route route VERB bracis-28410 131 5 capsules capsule NOUN bracis-28410 131 6 from from ADP bracis-28410 131 7 the the DET bracis-28410 131 8 previous previous ADJ bracis-28410 131 9 layer layer NOUN bracis-28410 131 10 to to ADP bracis-28410 131 11 capsules capsule NOUN bracis-28410 131 12 in in ADP bracis-28410 131 13 the the DET bracis-28410 131 14 next next ADJ bracis-28410 131 15 layer layer NOUN bracis-28410 131 16 within within ADP bracis-28410 131 17 a a DET bracis-28410 131 18 specific specific ADJ bracis-28410 131 19 spatial spatial ADJ bracis-28410 131 20 window window NOUN bracis-28410 131 21 . . PUNCT bracis-28410 132 1 the the DET bracis-28410 132 2 original original ADJ bracis-28410 132 3 algorithm algorithm NOUN bracis-28410 132 4 , , PUNCT bracis-28410 132 5 on on ADP bracis-28410 132 6 the the DET bracis-28410 132 7 other other ADJ bracis-28410 132 8 hand hand NOUN bracis-28410 132 9 , , PUNCT bracis-28410 132 10 directs direct VERB bracis-28410 132 11 the the DET bracis-28410 132 12 output output NOUN bracis-28410 132 13 of of ADP bracis-28410 132 14 all all DET bracis-28410 132 15 previous previous ADJ bracis-28410 132 16 layer layer NOUN bracis-28410 132 17 capsules capsule NOUN bracis-28410 132 18 to to ADP bracis-28410 132 19 all all DET bracis-28410 132 20 capsules capsule NOUN bracis-28410 132 21 in in ADP bracis-28410 132 22 the the DET bracis-28410 132 23 next next ADJ bracis-28410 132 24 layer layer NOUN bracis-28410 132 25 , , PUNCT bracis-28410 132 26 varying vary VERB bracis-28410 132 27 only only ADV bracis-28410 132 28 the the DET bracis-28410 132 29 routing routing NOUN bracis-28410 132 30 weight weight NOUN bracis-28410 132 31 . . PUNCT bracis-28410 133 1 second second ADJ bracis-28410 133 2 , , PUNCT bracis-28410 133 3 we we PRON bracis-28410 133 4 share share VERB bracis-28410 133 5 the the DET bracis-28410 133 6 transformation transformation NOUN bracis-28410 133 7 matrices matrix NOUN bracis-28410 133 8 among among ADP bracis-28410 133 9 all all DET bracis-28410 133 10 capsules capsule NOUN bracis-28410 133 11 of of ADP bracis-28410 133 12 the the DET bracis-28410 133 13 same same ADJ bracis-28410 133 14 type type NOUN bracis-28410 133 15 . . PUNCT bracis-28410 134 1 in in ADP bracis-28410 134 2 a a DET bracis-28410 134 3 later later ADJ bracis-28410 134 4 step step NOUN bracis-28410 134 5 , , PUNCT bracis-28410 134 6 we we PRON bracis-28410 134 7 decided decide VERB bracis-28410 134 8 to to PART bracis-28410 134 9 replace replace VERB bracis-28410 134 10 the the DET bracis-28410 134 11 initial initial ADJ bracis-28410 134 12 routing routing NOUN bracis-28410 134 13 algorithm algorithm NOUN bracis-28410 134 14 for for ADP bracis-28410 134 15 the the DET bracis-28410 134 16 no no PRON bracis-28410 134 17 - - PUNCT bracis-28410 134 18 routing routing NOUN bracis-28410 134 19 introduced introduce VERB bracis-28410 134 20 by by ADP bracis-28410 134 21 gu gu PROPN bracis-28410 134 22 et et PROPN bracis-28410 134 23 al al PROPN bracis-28410 134 24 . . PUNCT bracis-28410 134 25     SPACE bracis-28410 135 1 [ [ X bracis-28410 135 2 13 13 NUM bracis-28410 135 3 ] ] PUNCT bracis-28410 135 4 . . PUNCT bracis-28410 136 1 the the DET bracis-28410 136 2 authors author NOUN bracis-28410 136 3 argue argue VERB bracis-28410 136 4 that that SCONJ bracis-28410 136 5 the the DET bracis-28410 136 6 routing routing NOUN bracis-28410 136 7 procedure procedure NOUN bracis-28410 136 8 contributes contribute VERB bracis-28410 136 9 neither neither PRON bracis-28410 136 10 to to ADP bracis-28410 136 11 the the DET bracis-28410 136 12 generalization generalization NOUN bracis-28410 136 13 ability ability NOUN bracis-28410 136 14 nor nor CCONJ bracis-28410 136 15 to to ADP bracis-28410 136 16 the the DET bracis-28410 136 17 affine affine NOUN bracis-28410 136 18 robustness robustness NOUN bracis-28410 136 19 of of ADP bracis-28410 136 20 the the DET bracis-28410 136 21 capsnets capsnet NOUN bracis-28410 136 22 ; ; PUNCT bracis-28410 136 23 therefore therefore ADV bracis-28410 136 24 , , PUNCT bracis-28410 136 25 distinct distinct ADJ bracis-28410 136 26 ways way NOUN bracis-28410 136 27 to to PART bracis-28410 136 28 approximate approximate VERB bracis-28410 136 29 the the DET bracis-28410 136 30 coupling coupling NOUN bracis-28410 136 31 coefficients coefficient NOUN bracis-28410 136 32 do do AUX bracis-28410 136 33 not not PART bracis-28410 136 34 make make VERB bracis-28410 136 35 a a DET bracis-28410 136 36 significant significant ADJ bracis-28410 136 37 difference difference NOUN bracis-28410 136 38 since since SCONJ bracis-28410 136 39 they they PRON bracis-28410 136 40 will will AUX bracis-28410 136 41 be be AUX bracis-28410 136 42 learned learn VERB bracis-28410 136 43 implicitly implicitly ADV bracis-28410 136 44 . . PUNCT bracis-28410 137 1 in in ADP bracis-28410 137 2 the the DET bracis-28410 137 3 no no PRON bracis-28410 137 4 - - PUNCT bracis-28410 137 5 routing route VERB bracis-28410 137 6 approach approach NOUN bracis-28410 137 7 , , PUNCT bracis-28410 137 8 the the DET bracis-28410 137 9 iterative iterative NOUN bracis-28410 137 10 routing routing NOUN bracis-28410 137 11 procedure procedure NOUN bracis-28410 137 12 is be AUX bracis-28410 137 13 removed remove VERB bracis-28410 137 14 by by ADP bracis-28410 137 15 setting set VERB bracis-28410 137 16 all all DET bracis-28410 137 17 coupling couple VERB bracis-28410 137 18 coefficients coefficient NOUN bracis-28410 137 19 as as ADP bracis-28410 137 20 a a DET bracis-28410 137 21 constant constant ADJ bracis-28410 137 22 \(\frac{1}{m}\ \(\frac{1}{m}\ NOUN bracis-28410 137 23 ) ) PUNCT bracis-28410 137 24 , , PUNCT bracis-28410 137 25 where where SCONJ bracis-28410 137 26 m m NOUN bracis-28410 137 27 is be AUX bracis-28410 137 28 the the DET bracis-28410 137 29 number number NOUN bracis-28410 137 30 of of ADP bracis-28410 137 31 capsules capsule NOUN bracis-28410 137 32 in in ADP bracis-28410 137 33 the the DET bracis-28410 137 34 next next ADJ bracis-28410 137 35 layer layer NOUN bracis-28410 137 36 . . PUNCT bracis-28410 138 1 we we PRON bracis-28410 138 2 also also ADV bracis-28410 138 3 tried try VERB bracis-28410 138 4 different different ADJ bracis-28410 138 5 values value NOUN bracis-28410 138 6 for for ADP bracis-28410 138 7 the the DET bracis-28410 138 8 squashing squash VERB bracis-28410 138 9 constant constant ADJ bracis-28410 138 10 sq sq NOUN bracis-28410 138 11 used use VERB bracis-28410 138 12 in in ADP bracis-28410 138 13 the the DET bracis-28410 138 14 squashing squash VERB bracis-28410 138 15 function function NOUN bracis-28410 138 16 , , PUNCT bracis-28410 138 17 as as SCONJ bracis-28410 138 18 done do VERB bracis-28410 138 19 by by ADP bracis-28410 138 20 huang huang PROPN bracis-28410 138 21 and and CCONJ bracis-28410 138 22 zhou zhou PROPN bracis-28410 138 23     SPACE bracis-28410 139 1 [ [ X bracis-28410 139 2 20 20 NUM bracis-28410 139 3 ] ] PUNCT bracis-28410 139 4 . . PUNCT bracis-28410 140 1 the the DET bracis-28410 140 2 rdcbs rdcbs PROPN bracis-28410 140 3 are be AUX bracis-28410 140 4 sequentially sequentially ADV bracis-28410 140 5 connected connect VERB bracis-28410 140 6 , , PUNCT bracis-28410 140 7 each each PRON bracis-28410 140 8 having have VERB bracis-28410 140 9 a a DET bracis-28410 140 10 residual residual ADJ bracis-28410 140 11 connection connection NOUN bracis-28410 140 12 with with ADP bracis-28410 140 13 the the DET bracis-28410 140 14 block block NOUN bracis-28410 140 15 input input NOUN bracis-28410 140 16 , , PUNCT bracis-28410 140 17 followed follow VERB bracis-28410 140 18 by by ADP bracis-28410 140 19 a a DET bracis-28410 140 20 new new ADJ bracis-28410 140 21 convolutional convolutional ADJ bracis-28410 140 22 layer layer NOUN bracis-28410 140 23 , , PUNCT bracis-28410 140 24 and and CCONJ bracis-28410 140 25 the the DET bracis-28410 140 26 output output NOUN bracis-28410 140 27 of of ADP bracis-28410 140 28 that that DET bracis-28410 140 29 layer layer NOUN bracis-28410 140 30 has have VERB bracis-28410 140 31 a a DET bracis-28410 140 32 residual residual ADJ bracis-28410 140 33 connection connection NOUN bracis-28410 140 34 with with ADP bracis-28410 140 35 the the DET bracis-28410 140 36 output output NOUN bracis-28410 140 37 of of ADP bracis-28410 140 38 the the DET bracis-28410 140 39 first first ADJ bracis-28410 140 40 convolutional convolutional ADJ bracis-28410 140 41 layer layer NOUN bracis-28410 140 42 . . PUNCT bracis-28410 141 1 we we PRON bracis-28410 141 2 use use VERB bracis-28410 141 3 residual residual ADJ bracis-28410 141 4 connections connection NOUN bracis-28410 141 5 , , PUNCT bracis-28410 141 6 identified identify VERB bracis-28410 141 7 by by ADP bracis-28410 141 8 the the DET bracis-28410 141 9 symbol symbol NOUN bracis-28410 141 10 in in ADP bracis-28410 141 11 figs fig NOUN bracis-28410 141 12 . . PUNCT bracis-28410 141 13     SPACE bracis-28410 141 14 1 1 NUM bracis-28410 141 15 and and CCONJ bracis-28410 141 16     SPACE bracis-28410 141 17 2 2 NUM bracis-28410 141 18 , , PUNCT bracis-28410 141 19 for for ADP bracis-28410 141 20 the the DET bracis-28410 141 21 following follow VERB bracis-28410 141 22 purposes purpose NOUN bracis-28410 141 23 : : PUNCT bracis-28410 141 24 they they PRON bracis-28410 141 25 avoid avoid VERB bracis-28410 141 26 the the DET bracis-28410 141 27 problem problem NOUN bracis-28410 141 28 of of ADP bracis-28410 141 29 vanishing vanish VERB bracis-28410 141 30 gradients gradient NOUN bracis-28410 141 31 ( ( PUNCT bracis-28410 141 32 it it PRON bracis-28410 141 33 becomes become VERB bracis-28410 141 34 zero zero NUM bracis-28410 141 35 ) ) PUNCT bracis-28410 141 36 by by ADP bracis-28410 141 37 introducing introduce VERB bracis-28410 141 38 shorter short ADJ bracis-28410 141 39 paths path NOUN bracis-28410 141 40 , , PUNCT bracis-28410 141 41 which which PRON bracis-28410 141 42 can can AUX bracis-28410 141 43 take take VERB bracis-28410 141 44 the the DET bracis-28410 141 45 gradient gradient NOUN bracis-28410 141 46 over over ADP bracis-28410 141 47 the the DET bracis-28410 141 48 entire entire ADJ bracis-28410 141 49 length length NOUN bracis-28410 141 50 of of ADP bracis-28410 141 51 very very ADV bracis-28410 141 52 deep deep ADJ bracis-28410 141 53 networks network NOUN bracis-28410 141 54 , , PUNCT bracis-28410 141 55 as as SCONJ bracis-28410 141 56 demonstrated demonstrate VERB bracis-28410 141 57 by by ADP bracis-28410 141 58     SPACE bracis-28410 141 59 veit veit PROPN bracis-28410 141 60 et et PROPN bracis-28410 141 61 al al PROPN bracis-28410 141 62 . . PUNCT bracis-28410 141 63     SPACE bracis-28410 142 1 [ [ X bracis-28410 142 2 52 52 NUM bracis-28410 142 3 ] ] PUNCT bracis-28410 142 4 ; ; PUNCT bracis-28410 142 5 and and CCONJ bracis-28410 142 6 the the DET bracis-28410 142 7 use use NOUN bracis-28410 142 8 of of ADP bracis-28410 142 9 residual residual ADJ bracis-28410 142 10 connections connection NOUN bracis-28410 142 11 seems seem VERB bracis-28410 142 12 to to PART bracis-28410 142 13 greatly greatly ADV bracis-28410 142 14 improve improve VERB bracis-28410 142 15 training training NOUN bracis-28410 142 16 speed speed NOUN bracis-28410 142 17     SPACE bracis-28410 143 1 [ [ X bracis-28410 143 2 48 48 NUM bracis-28410 143 3 ] ] PUNCT bracis-28410 143 4 . . PUNCT bracis-28410 144 1 at at ADP bracis-28410 144 2 the the DET bracis-28410 144 3 end end NOUN bracis-28410 144 4 of of ADP bracis-28410 144 5 the the DET bracis-28410 144 6 model model NOUN bracis-28410 144 7 , , PUNCT bracis-28410 144 8 there there PRON bracis-28410 144 9 is be VERB bracis-28410 144 10 a a DET bracis-28410 144 11 network network NOUN bracis-28410 144 12 used use VERB bracis-28410 144 13 to to PART bracis-28410 144 14 upscale upscale NOUN bracis-28410 144 15 , , PUNCT bracis-28410 144 16 called call VERB bracis-28410 144 17 up up ADV bracis-28410 144 18 - - PUNCT bracis-28410 144 19 sampling sample VERB bracis-28410 144 20 network network NOUN bracis-28410 144 21 ( ( PUNCT bracis-28410 144 22 upnet upnet NOUN bracis-28410 144 23 ) ) PUNCT bracis-28410 144 24 ( ( PUNCT bracis-28410 144 25 see see VERB bracis-28410 144 26 fig fig NOUN bracis-28410 144 27 . . PUNCT bracis-28410 144 28     SPACE bracis-28410 144 29 3 3 NUM bracis-28410 144 30 ) ) PUNCT bracis-28410 144 31 . . PUNCT bracis-28410 145 1 following follow VERB bracis-28410 145 2 previous previous ADJ bracis-28410 145 3 works work NOUN bracis-28410 145 4     SPACE bracis-28410 146 1 [ [ X bracis-28410 146 2 33 33 NUM bracis-28410 146 3 , , PUNCT bracis-28410 146 4 64 64 NUM bracis-28410 146 5 ] ] PUNCT bracis-28410 146 6 and and CCONJ bracis-28410 146 7 several several ADJ bracis-28410 146 8 participants participant NOUN bracis-28410 146 9 of of ADP bracis-28410 146 10 the the DET bracis-28410 146 11 ntire ntire NOUN bracis-28410 146 12 2017 2017 NUM bracis-28410 146 13     SPACE bracis-28410 147 1 [ [ X bracis-28410 147 2 50 50 NUM bracis-28410 147 3 ] ] PUNCT bracis-28410 147 4 and and CCONJ bracis-28410 147 5 ntire ntire NOUN bracis-28410 147 6 2018 2018 NUM bracis-28410 147 7     SPACE bracis-28410 148 1 [ [ X bracis-28410 148 2 51 51 NUM bracis-28410 148 3 ] ] PUNCT bracis-28410 148 4 competitions competition NOUN bracis-28410 148 5 , , PUNCT bracis-28410 148 6 the the DET bracis-28410 148 7 upnet upnet NOUN bracis-28410 148 8 is be AUX bracis-28410 148 9 composed compose VERB bracis-28410 148 10 of of ADP bracis-28410 148 11 subpixel subpixel ADJ bracis-28410 148 12 convolutions convolution NOUN bracis-28410 148 13     SPACE bracis-28410 149 1 [ [ X bracis-28410 149 2 44 44 NUM bracis-28410 149 3 ] ] PUNCT bracis-28410 149 4 . . PUNCT bracis-28410 150 1 the the DET bracis-28410 150 2 upnet upnet NOUN bracis-28410 150 3 allows allow VERB bracis-28410 150 4 the the DET bracis-28410 150 5 network network NOUN bracis-28410 150 6 to to PART bracis-28410 150 7 implicitly implicitly ADV bracis-28410 150 8 learn learn VERB bracis-28410 150 9 the the DET bracis-28410 150 10 process process NOUN bracis-28410 150 11 required require VERB bracis-28410 150 12 to to PART bracis-28410 150 13 generate generate VERB bracis-28410 150 14 the the DET bracis-28410 150 15 larger large ADJ bracis-28410 150 16 version version NOUN bracis-28410 150 17 by by ADP bracis-28410 150 18 adding add VERB bracis-28410 150 19 the the DET bracis-28410 150 20 lr lr PROPN bracis-28410 150 21 space space NOUN bracis-28410 150 22 feature feature NOUN bracis-28410 150 23 maps map NOUN bracis-28410 150 24 and and CCONJ bracis-28410 150 25 creating create VERB bracis-28410 150 26 the the DET bracis-28410 150 27 sr sr PROPN bracis-28410 150 28 image image NOUN bracis-28410 150 29 in in ADP bracis-28410 150 30 a a DET bracis-28410 150 31 single single ADJ bracis-28410 150 32 step step NOUN bracis-28410 150 33 , , PUNCT bracis-28410 150 34 saving save VERB bracis-28410 150 35 memory memory NOUN bracis-28410 150 36 and and CCONJ bracis-28410 150 37 processing processing NOUN bracis-28410 150 38 . . PUNCT bracis-28410 151 1 we we PRON bracis-28410 151 2 prefer prefer VERB bracis-28410 151 3 this this DET bracis-28410 151 4 method method NOUN bracis-28410 151 5 over over ADP bracis-28410 151 6 deconvolution deconvolution NOUN bracis-28410 151 7 since since SCONJ bracis-28410 151 8 it it PRON bracis-28410 151 9 naturally naturally ADV bracis-28410 151 10 avoids avoid VERB bracis-28410 151 11 checkerboard checkerboard NOUN bracis-28410 151 12 artifacts artifact NOUN bracis-28410 151 13 , , PUNCT bracis-28410 151 14 which which PRON bracis-28410 151 15 with with ADP bracis-28410 151 16 deconvolution deconvolution NOUN bracis-28410 151 17 must must AUX bracis-28410 151 18 be be AUX bracis-28410 151 19 done do VERB bracis-28410 151 20 using use VERB bracis-28410 151 21 a a DET bracis-28410 151 22 kernel kernel NOUN bracis-28410 151 23 size size NOUN bracis-28410 151 24 that that PRON bracis-28410 151 25 is be AUX bracis-28410 151 26 divisible divisible ADJ bracis-28410 151 27 by by ADP bracis-28410 151 28 stride stride NOUN bracis-28410 151 29 to to PART bracis-28410 151 30 avoid avoid VERB bracis-28410 151 31 the the DET bracis-28410 151 32 overlapping overlap VERB bracis-28410 151 33 problem problem NOUN bracis-28410 151 34 as as SCONJ bracis-28410 151 35 demonstrated demonstrate VERB bracis-28410 151 36 by by ADP bracis-28410 151 37 odena odena NOUN bracis-28410 151 38 et et PROPN bracis-28410 151 39 al al PROPN bracis-28410 151 40 . . PUNCT bracis-28410 151 41     SPACE bracis-28410 152 1 [ [ X bracis-28410 152 2 38 38 NUM bracis-28410 152 3 ] ] PUNCT bracis-28410 152 4 . . PUNCT bracis-28410 153 1 besides besides SCONJ bracis-28410 153 2 , , PUNCT bracis-28410 153 3 upnet upnet NOUN bracis-28410 153 4 has have VERB bracis-28410 153 5 a a DET bracis-28410 153 6 considerably considerably ADV bracis-28410 153 7 lower low ADJ bracis-28410 153 8 computational computational ADJ bracis-28410 153 9 cost cost NOUN bracis-28410 153 10 , , PUNCT bracis-28410 153 11 becoming become VERB bracis-28410 153 12 \(\log \(\log NOUN bracis-28410 153 13 _ _ DET bracis-28410 153 14 { { PUNCT bracis-28410 153 15 2}r^{2}\ 2}r^{2}\ NUM bracis-28410 153 16 ) ) PUNCT bracis-28410 153 17 times time NOUN bracis-28410 153 18 faster fast ADV bracis-28410 153 19 during during ADP bracis-28410 153 20 training training NOUN bracis-28410 153 21     SPACE bracis-28410 154 1 [ [ X bracis-28410 154 2 44 44 NUM bracis-28410 154 3 ] ] PUNCT bracis-28410 154 4 . . PUNCT bracis-28410 155 1 fig fig NOUN bracis-28410 155 2 . . PUNCT bracis-28410 156 1 3 3 X bracis-28410 156 2 . . X bracis-28410 156 3 upnet upnet NOUN bracis-28410 156 4 diagram diagram NOUN bracis-28410 156 5 . . PUNCT bracis-28410 157 1 full full ADJ bracis-28410 157 2 size size NOUN bracis-28410 157 3 image image NOUN bracis-28410 157 4 loss loss NOUN bracis-28410 157 5 functions function NOUN bracis-28410 157 6 . . PUNCT bracis-28410 158 1 during during ADP bracis-28410 158 2 training training NOUN bracis-28410 158 3 , , PUNCT bracis-28410 158 4 we we PRON bracis-28410 158 5 evaluated evaluate VERB bracis-28410 158 6 loss loss NOUN bracis-28410 158 7 functions function NOUN bracis-28410 158 8 commonly commonly ADV bracis-28410 158 9 used use VERB bracis-28410 158 10 in in ADP bracis-28410 158 11 super super ADJ bracis-28410 158 12 - - ADJ bracis-28410 158 13 resolution resolution ADJ bracis-28410 158 14 problems problem NOUN bracis-28410 158 15     SPACE bracis-28410 159 1 [ [ X bracis-28410 159 2 56 56 NUM bracis-28410 159 3 ] ] PUNCT bracis-28410 159 4 . . PUNCT bracis-28410 160 1 due due ADP bracis-28410 160 2 to to ADP bracis-28410 160 3 its its PRON bracis-28410 160 4 simplicity simplicity NOUN bracis-28410 160 5 and and CCONJ bracis-28410 160 6 effectiveness effectiveness NOUN bracis-28410 160 7 , , PUNCT bracis-28410 160 8 the the DET bracis-28410 160 9 first first ADJ bracis-28410 160 10 loss loss NOUN bracis-28410 160 11 function function NOUN bracis-28410 160 12 we we PRON bracis-28410 160 13 assess assess VERB bracis-28410 160 14 is be AUX bracis-28410 160 15 the the DET bracis-28410 160 16 l1 l1 PROPN bracis-28410 160 17 . . PUNCT bracis-28410 161 1 previous previous ADJ bracis-28410 161 2 studies study NOUN bracis-28410 161 3     SPACE bracis-28410 162 1 [ [ X bracis-28410 162 2 33 33 NUM bracis-28410 162 3 , , PUNCT bracis-28410 162 4 65 65 NUM bracis-28410 162 5 ] ] PUNCT bracis-28410 162 6 showed show VERB bracis-28410 162 7 that that SCONJ bracis-28410 162 8 a a DET bracis-28410 162 9 network network NOUN bracis-28410 162 10 trained train VERB bracis-28410 162 11 with with ADP bracis-28410 162 12 l1 l1 PROPN bracis-28410 162 13 has have AUX bracis-28410 162 14 achieved achieve VERB bracis-28410 162 15 superior superior ADJ bracis-28410 162 16 results result NOUN bracis-28410 162 17 compared compare VERB bracis-28410 162 18 to to ADP bracis-28410 162 19 the the DET bracis-28410 162 20 same same ADJ bracis-28410 162 21 network network NOUN bracis-28410 162 22 trained train VERB bracis-28410 162 23 with with ADP bracis-28410 162 24 l2 l2 NOUN bracis-28410 162 25 . . PUNCT bracis-28410 163 1 still still ADV bracis-28410 163 2 in in ADP bracis-28410 163 3 the the DET bracis-28410 163 4 work work NOUN bracis-28410 163 5 of of ADP bracis-28410 163 6 zhao zhao PROPN bracis-28410 163 7 et et PROPN bracis-28410 163 8 al al PROPN bracis-28410 163 9 . . PUNCT bracis-28410 163 10     SPACE bracis-28410 164 1 [ [ X bracis-28410 164 2 65 65 NUM bracis-28410 164 3 ] ] PUNCT bracis-28410 164 4 , , PUNCT bracis-28410 164 5 the the DET bracis-28410 164 6 idea idea NOUN bracis-28410 164 7 of of ADP bracis-28410 164 8 using use VERB bracis-28410 164 9 indices index NOUN bracis-28410 164 10 based base VERB bracis-28410 164 11 on on ADP bracis-28410 164 12 the the DET bracis-28410 164 13 structural structural ADJ bracis-28410 164 14 similarity similarity NOUN bracis-28410 164 15 index index NOUN bracis-28410 164 16 ( ( PUNCT bracis-28410 164 17 ssim ssim NOUN bracis-28410 164 18 ) ) PUNCT bracis-28410 164 19     SPACE bracis-28410 165 1 [ [ X bracis-28410 165 2 57 57 NUM bracis-28410 165 3 ] ] PUNCT bracis-28410 165 4 is be AUX bracis-28410 165 5 introduced introduce VERB bracis-28410 165 6 for for ADP bracis-28410 165 7 training train VERB bracis-28410 165 8 neural neural ADJ bracis-28410 165 9 networks network NOUN bracis-28410 165 10 . . PUNCT bracis-28410 166 1 as as SCONJ bracis-28410 166 2 previously previously ADV bracis-28410 166 3 noted note VERB bracis-28410 166 4 by by ADP bracis-28410 166 5 dong dong PROPN bracis-28410 166 6 et et PROPN bracis-28410 166 7 al al PROPN bracis-28410 166 8 . . PUNCT bracis-28410 166 9     SPACE bracis-28410 167 1 [ [ X bracis-28410 167 2 11 11 NUM bracis-28410 167 3 ] ] PUNCT bracis-28410 167 4 , , PUNCT bracis-28410 167 5 if if SCONJ bracis-28410 167 6 a a DET bracis-28410 167 7 metric metric NOUN bracis-28410 167 8 based base VERB bracis-28410 167 9 on on ADP bracis-28410 167 10 visual visual ADJ bracis-28410 167 11 perception perception NOUN bracis-28410 167 12 is be AUX bracis-28410 167 13 used use VERB bracis-28410 167 14 during during ADP bracis-28410 167 15 training training NOUN bracis-28410 167 16 , , PUNCT bracis-28410 167 17 the the DET bracis-28410 167 18 network network NOUN bracis-28410 167 19 can can AUX bracis-28410 167 20 adapt adapt VERB bracis-28410 167 21 to to ADP bracis-28410 167 22 it it PRON bracis-28410 167 23 . . PUNCT bracis-28410 168 1 the the DET bracis-28410 168 2 ssim ssim NOUN bracis-28410 168 3 is be AUX bracis-28410 168 4 calculated calculate VERB bracis-28410 168 5 as as ADP bracis-28410 168 6 : : PUNCT bracis-28410 168 7 $ $ SYM bracis-28410 168 8 $ $ SYM bracis-28410 168 9 \begin{aligned \begin{aligne VERB bracis-28410 168 10 } } PUNCT bracis-28410 168 11 \text \text ADV bracis-28410 168 12 { { PUNCT bracis-28410 168 13 ssim}(p ssim}(p X bracis-28410 168 14 ) ) PUNCT bracis-28410 169 1 = = NUM bracis-28410 169 2 \frac{2 \frac{2 ADP bracis-28410 169 3 \mu \mu PROPN bracis-28410 169 4 _ _ PROPN bracis-28410 169 5 { { PUNCT bracis-28410 169 6 x x NOUN bracis-28410 169 7 } } PUNCT bracis-28410 169 8 \mu \mu NOUN bracis-28410 169 9 _ _ SYM bracis-28410 169 10 { { PUNCT bracis-28410 169 11 y}+c_{1}}{\mu y}+c_{1}}{\mu NUM bracis-28410 169 12 _ _ PUNCT bracis-28410 169 13 { { PUNCT bracis-28410 169 14 x}^{2}+\mu x}^{2}+\mu X bracis-28410 170 1 _ _ PRON bracis-28410 170 2 { { PUNCT bracis-28410 170 3 y}^{2}+c_{1 y}^{2}+c_{1 NOUN bracis-28410 170 4 } } PUNCT bracis-28410 170 5 } } PUNCT bracis-28410 170 6 \cdot \cdot PROPN bracis-28410 170 7 \frac{2 \frac{2 ADP bracis-28410 170 8 \sigma \sigma PROPN bracis-28410 170 9 _ _ PUNCT bracis-28410 170 10 { { PUNCT bracis-28410 170 11 x x X bracis-28410 170 12 y}+c_{2}}{\sigma y}+c_{2}}{\sigma SYM bracis-28410 170 13 _ _ PROPN bracis-28410 170 14 { { PUNCT bracis-28410 170 15 x}^{2}+\sigma x}^{2}+\sigma X bracis-28410 170 16 _ _ PUNCT bracis-28410 170 17 { { PUNCT bracis-28410 170 18 y}^{2}+c_{2 y}^{2}+c_{2 NOUN bracis-28410 170 19 } } PUNCT bracis-28410 170 20 } } PUNCT bracis-28410 170 21 = = SYM bracis-28410 170 22 l(p l(p NOUN bracis-28410 170 23 ) ) PUNCT bracis-28410 170 24 \cdot \cdot NOUN bracis-28410 170 25 cs(p cs(p NUM bracis-28410 170 26 ) ) PUNCT bracis-28410 170 27 \end{aligned}$$ \end{aligned}$$ X bracis-28410 170 28 ( ( PUNCT bracis-28410 170 29 1 1 NUM bracis-28410 170 30 ) ) PUNCT bracis-28410 170 31 where where SCONJ bracis-28410 170 32 \(\mu \(\mu NOUN bracis-28410 170 33 _ _ PUNCT bracis-28410 170 34 { { PUNCT bracis-28410 170 35 x}\ x}\ PROPN bracis-28410 170 36 ) ) PUNCT bracis-28410 170 37 and and CCONJ bracis-28410 170 38 \(\mu \(\mu VERB bracis-28410 170 39 _ _ PUNCT bracis-28410 170 40 { { PUNCT bracis-28410 170 41 y}\ y}\ NOUN bracis-28410 170 42 ) ) PUNCT bracis-28410 170 43 are be AUX bracis-28410 170 44 the the DET bracis-28410 170 45 average average ADJ bracis-28410 170 46 pixel pixel NOUN bracis-28410 170 47 values value NOUN bracis-28410 170 48 in in ADP bracis-28410 170 49 the the DET bracis-28410 170 50 sr sr PROPN bracis-28410 170 51 and and CCONJ bracis-28410 170 52 hr hr NOUN bracis-28410 170 53 patches patch NOUN bracis-28410 170 54 , , PUNCT bracis-28410 170 55 respectively respectively ADV bracis-28410 170 56 , , PUNCT bracis-28410 170 57 \(\sigma \(\sigma NOUN bracis-28410 170 58 _ _ PRON bracis-28410 170 59 { { PUNCT bracis-28410 170 60 x}\ x}\ PROPN bracis-28410 170 61 ) ) PUNCT bracis-28410 170 62 and and CCONJ bracis-28410 170 63 \(\sigma \(\sigma NOUN bracis-28410 170 64 _ _ PUNCT bracis-28410 170 65 { { PUNCT bracis-28410 170 66 y}\ y}\ NOUN bracis-28410 170 67 ) ) PUNCT bracis-28410 170 68 are be AUX bracis-28410 170 69 the the DET bracis-28410 170 70 standard standard ADJ bracis-28410 170 71 deviations deviation NOUN bracis-28410 170 72 of of ADP bracis-28410 170 73 the the DET bracis-28410 170 74 same same ADJ bracis-28410 170 75 patches patch NOUN bracis-28410 170 76 , , PUNCT bracis-28410 170 77 \(\sigma \(\sigma X bracis-28410 170 78 _ _ PRON bracis-28410 170 79 { { PUNCT bracis-28410 170 80 x x SYM bracis-28410 170 81 y}\ y}\ NOUN bracis-28410 170 82 ) ) PUNCT bracis-28410 170 83 is be AUX bracis-28410 170 84 the the DET bracis-28410 170 85 covariance covariance NOUN bracis-28410 170 86 between between ADP bracis-28410 170 87 them they PRON bracis-28410 170 88 , , PUNCT bracis-28410 170 89 and and CCONJ bracis-28410 170 90 \(c_{1}\ \(c_{1}\ PROPN bracis-28410 170 91 ) ) PUNCT bracis-28410 170 92 and and CCONJ bracis-28410 170 93 \(c_{2}\ \(c_{2}\ NOUN bracis-28410 170 94 ) ) PUNCT bracis-28410 171 1 are be AUX bracis-28410 171 2 constants constant NOUN bracis-28410 171 3 added add VERB bracis-28410 171 4 to to PART bracis-28410 171 5 avoid avoid VERB bracis-28410 171 6 instabilities instability NOUN bracis-28410 171 7 when when SCONJ bracis-28410 171 8 the the DET bracis-28410 171 9 values value NOUN bracis-28410 171 10 of of ADP bracis-28410 171 11 \(\mu \(\mu PROPN bracis-28410 171 12 _ _ PUNCT bracis-28410 171 13 { { PUNCT bracis-28410 171 14 x}^{2}+\mu x}^{2}+\mu X bracis-28410 172 1 _ _ PUNCT bracis-28410 172 2 { { PUNCT bracis-28410 172 3 y}^{2}\ y}^{2}\ NOUN bracis-28410 172 4 ) ) PUNCT bracis-28410 172 5 and and CCONJ bracis-28410 172 6 \(\sigma \(\sigma NOUN bracis-28410 172 7 _ _ PRON bracis-28410 172 8 { { PUNCT bracis-28410 172 9 x}^{2}+\sigma x}^{2}+\sigma X bracis-28410 172 10 _ _ PUNCT bracis-28410 172 11 { { PUNCT bracis-28410 172 12 y}^{2}\ y}^{2}\ NOUN bracis-28410 172 13 ) ) PUNCT bracis-28410 172 14 are be AUX bracis-28410 172 15 very very ADV bracis-28410 172 16 close close ADJ bracis-28410 172 17 to to ADP bracis-28410 172 18 0 0 NUM bracis-28410 172 19 . . PUNCT bracis-28410 173 1 the the DET bracis-28410 173 2 l(p l(p NOUN bracis-28410 173 3 ) ) PUNCT bracis-28410 173 4 part part NOUN bracis-28410 173 5 of of ADP bracis-28410 173 6 the the DET bracis-28410 173 7 equation equation NOUN bracis-28410 173 8 calculates calculate VERB bracis-28410 173 9 the the DET bracis-28410 173 10 comparison comparison NOUN bracis-28410 173 11 between between ADP bracis-28410 173 12 clipping clip VERB bracis-28410 173 13 luminances luminance NOUN bracis-28410 173 14 , , PUNCT bracis-28410 173 15 while while SCONJ bracis-28410 173 16 the the DET bracis-28410 173 17 comparison comparison NOUN bracis-28410 173 18 between between ADP bracis-28410 173 19 their their PRON bracis-28410 173 20 contrasts contrast NOUN bracis-28410 173 21 and and CCONJ bracis-28410 173 22 structures structure NOUN bracis-28410 173 23 is be AUX bracis-28410 173 24 calculated calculate VERB bracis-28410 173 25 by by ADP bracis-28410 173 26 cs(p cs(p NOUN bracis-28410 173 27 ) ) PUNCT bracis-28410 173 28 . . PUNCT bracis-28410 174 1 since since SCONJ bracis-28410 174 2 the the DET bracis-28410 174 3 highest high ADJ bracis-28410 174 4 possible possible ADJ bracis-28410 174 5 value value NOUN bracis-28410 174 6 for for ADP bracis-28410 174 7 ssim ssim NOUN bracis-28410 174 8 is be AUX bracis-28410 174 9 1 1 NUM bracis-28410 174 10 , , PUNCT bracis-28410 174 11 and and CCONJ bracis-28410 174 12 because because SCONJ bracis-28410 174 13 training train VERB bracis-28410 174 14 a a DET bracis-28410 174 15 neural neural ADJ bracis-28410 174 16 network network NOUN bracis-28410 174 17 usually usually ADV bracis-28410 174 18 aims aim VERB bracis-28410 174 19 to to PART bracis-28410 174 20 minimize minimize VERB bracis-28410 174 21 the the DET bracis-28410 174 22 loss loss NOUN bracis-28410 174 23 function function NOUN bracis-28410 174 24 , , PUNCT bracis-28410 174 25 we we PRON bracis-28410 174 26 can can AUX bracis-28410 174 27 define define VERB bracis-28410 174 28 the the DET bracis-28410 174 29 \(\mathcal \(\mathcal ADJ bracis-28410 174 30 { { PUNCT bracis-28410 174 31 l}^{\text l}^{\text ADJ bracis-28410 174 32 { { PUNCT bracis-28410 174 33 ssim}}\ ssim}}\ NOUN bracis-28410 174 34 ) ) PUNCT bracis-28410 174 35 function function NOUN bracis-28410 174 36 as as ADP bracis-28410 174 37 : : PUNCT bracis-28410 174 38 $ $ SYM bracis-28410 174 39 $ $ SYM bracis-28410 174 40 \begin{aligned \begin{aligne VERB bracis-28410 174 41 } } PUNCT bracis-28410 174 42 \mathcal \mathcal ADJ bracis-28410 174 43 { { PUNCT bracis-28410 174 44 l}^{\text l}^{\text X bracis-28410 174 45 { { PUNCT bracis-28410 174 46 ssim}}(p)=1-\text ssim}}(p)=1-\text PROPN bracis-28410 174 47 { { PUNCT bracis-28410 174 48 ssim}(\tilde{p ssim}(\tilde{p NOUN bracis-28410 174 49 } } PUNCT bracis-28410 174 50 ) ) PUNCT bracis-28410 174 51 , , PUNCT bracis-28410 174 52 \end{aligned}$$ \end{aligned}$$ X bracis-28410 174 53 ( ( PUNCT bracis-28410 174 54 2 2 NUM bracis-28410 174 55 ) ) PUNCT bracis-28410 174 56 in in ADP bracis-28410 174 57 which which PRON bracis-28410 174 58 \(\tilde{p}\ \(\tilde{p}\ NOUN bracis-28410 174 59 ) ) PUNCT bracis-28410 174 60 is be AUX bracis-28410 174 61 the the DET bracis-28410 174 62 central central ADJ bracis-28410 174 63 pixel pixel NOUN bracis-28410 174 64 of of ADP bracis-28410 174 65 patch patch PROPN bracis-28410 175 1 p. p. PROPN bracis-28410 175 2 the the DET bracis-28410 175 3 best good ADJ bracis-28410 175 4 performance performance NOUN bracis-28410 175 5 in in ADP bracis-28410 175 6 the the DET bracis-28410 175 7 work work NOUN bracis-28410 175 8 of of ADP bracis-28410 175 9 zhao zhao PROPN bracis-28410 175 10 et et PROPN bracis-28410 175 11 al al PROPN bracis-28410 175 12 . . PUNCT bracis-28410 175 13     SPACE bracis-28410 176 1 [ [ X bracis-28410 176 2 65 65 NUM bracis-28410 176 3 ] ] PUNCT bracis-28410 176 4 was be AUX bracis-28410 176 5 obtained obtain VERB bracis-28410 176 6 by by ADP bracis-28410 176 7 combining combine VERB bracis-28410 176 8 l1 l1 PROPN bracis-28410 176 9 and and CCONJ bracis-28410 176 10 the the DET bracis-28410 176 11 multi multi ADJ bracis-28410 176 12 - - ADJ bracis-28410 176 13 scale scale ADJ bracis-28410 176 14 structural structural ADJ bracis-28410 176 15 similarity similarity NOUN bracis-28410 176 16 index index NOUN bracis-28410 176 17 ( ( PUNCT bracis-28410 176 18 ms ms NOUN bracis-28410 176 19 - - PUNCT bracis-28410 176 20 ssim ssim NOUN bracis-28410 176 21 ) ) PUNCT bracis-28410 176 22 shown show VERB bracis-28410 176 23 in in ADP bracis-28410 176 24 eq eq PROPN bracis-28410 176 25 . . PUNCT bracis-28410 176 26     SPACE bracis-28410 176 27 4 4 NUM bracis-28410 176 28 weighing weigh VERB bracis-28410 176 29 0.16 0.16 NUM bracis-28410 176 30 and and CCONJ bracis-28410 176 31 0.84 0.84 NUM bracis-28410 176 32 , , PUNCT bracis-28410 176 33 respectively respectively ADV bracis-28410 176 34 . . PUNCT bracis-28410 177 1 the the DET bracis-28410 177 2 authors author NOUN bracis-28410 177 3 argue argue VERB bracis-28410 177 4 that that SCONJ bracis-28410 177 5 ms ms ADJ bracis-28410 177 6 - - PUNCT bracis-28410 177 7 ssim ssim NOUN bracis-28410 177 8 preserves preserve VERB bracis-28410 177 9 contrast contrast VERB bracis-28410 177 10 in in ADP bracis-28410 177 11 high high ADJ bracis-28410 177 12 - - PUNCT bracis-28410 177 13 frequency frequency NOUN bracis-28410 177 14 regions region NOUN bracis-28410 177 15 , , PUNCT bracis-28410 177 16 while while SCONJ bracis-28410 177 17 the the DET bracis-28410 177 18 l1 l1 PROPN bracis-28410 177 19 preserves preserve VERB bracis-28410 177 20 color color NOUN bracis-28410 177 21 and and CCONJ bracis-28410 177 22 brightness brightness NOUN bracis-28410 177 23 regardless regardless ADV bracis-28410 177 24 of of ADP bracis-28410 177 25 the the DET bracis-28410 177 26 local local ADJ bracis-28410 177 27 structure structure NOUN bracis-28410 177 28 . . PUNCT bracis-28410 178 1 the the DET bracis-28410 178 2 ms ms ADJ bracis-28410 178 3 - - PUNCT bracis-28410 178 4 ssim ssim NOUN bracis-28410 178 5 value value NOUN bracis-28410 178 6 is be AUX bracis-28410 178 7 obtained obtain VERB bracis-28410 178 8 by by ADP bracis-28410 178 9 combining combine VERB bracis-28410 178 10 measurements measurement NOUN bracis-28410 178 11 at at ADP bracis-28410 178 12 different different ADJ bracis-28410 178 13 scales scale NOUN bracis-28410 178 14 using use VERB bracis-28410 178 15 the the DET bracis-28410 178 16 equation equation NOUN bracis-28410 178 17 : : PUNCT bracis-28410 178 18 $ $ SYM bracis-28410 178 19 $ $ SYM bracis-28410 178 20 \begin{aligned \begin{aligne VERB bracis-28410 178 21 } } PUNCT bracis-28410 178 22 \text \text ADP bracis-28410 178 23 { { PUNCT bracis-28410 178 24 ms ms NOUN bracis-28410 178 25 - - PUNCT bracis-28410 178 26 ssim}(p)=l_{m}^{\alpha ssim}(p)=l_{m}^{\alpha NOUN bracis-28410 178 27 } } PUNCT bracis-28410 178 28 ( ( PUNCT bracis-28410 178 29 p p NOUN bracis-28410 178 30 ) ) PUNCT bracis-28410 178 31 \cdot \cdot NOUN bracis-28410 178 32 \prod \prod PROPN bracis-28410 178 33 _ _ PROPN bracis-28410 178 34 { { PUNCT bracis-28410 178 35 j=1}^{m j=1}^{m PROPN bracis-28410 178 36 } } PUNCT bracis-28410 178 37 cs_{j}^{\beta cs_{j}^{\beta PROPN bracis-28410 178 38 _ _ PROPN bracis-28410 178 39 { { PUNCT bracis-28410 178 40 j}}(p j}}(p NOUN bracis-28410 178 41 ) ) PUNCT bracis-28410 178 42 , , PUNCT bracis-28410 178 43 \end{aligned}$$ \end{aligned}$$ X bracis-28410 178 44 ( ( PUNCT bracis-28410 178 45 3 3 NUM bracis-28410 178 46 ) ) PUNCT bracis-28410 178 47 where where SCONJ bracis-28410 178 48 scales scale NOUN bracis-28410 178 49 are be AUX bracis-28410 178 50 used use VERB bracis-28410 178 51 ranging range VERB bracis-28410 178 52 from from ADP bracis-28410 178 53 1 1 NUM bracis-28410 178 54 ( ( PUNCT bracis-28410 178 55 original original ADJ bracis-28410 178 56 image image NOUN bracis-28410 178 57 ) ) PUNCT bracis-28410 178 58 to to ADP bracis-28410 178 59 m m PROPN bracis-28410 178 60 ( ( PUNCT bracis-28410 178 61 the the DET bracis-28410 178 62 largest large ADJ bracis-28410 178 63 scale scale NOUN bracis-28410 178 64 used use VERB bracis-28410 178 65 ) ) PUNCT bracis-28410 178 66 , , PUNCT bracis-28410 178 67 reducing reduce VERB bracis-28410 178 68 the the DET bracis-28410 178 69 image image NOUN bracis-28410 178 70 by by ADP bracis-28410 178 71 a a DET bracis-28410 178 72 factor factor NOUN bracis-28410 178 73 of of ADP bracis-28410 178 74 2 2 NUM bracis-28410 178 75 every every DET bracis-28410 178 76 iteration iteration NOUN bracis-28410 178 77 ; ; PUNCT bracis-28410 178 78 \(l_{m}\ \(l_{m}\ NOUN bracis-28410 178 79 ) ) PUNCT bracis-28410 178 80 and and CCONJ bracis-28410 178 81 \(cs_{j}\ \(cs_{j}\ NOUN bracis-28410 178 82 ) ) PUNCT bracis-28410 178 83 are be AUX bracis-28410 178 84 the the DET bracis-28410 178 85 same same ADJ bracis-28410 178 86 terms term NOUN bracis-28410 178 87 as as SCONJ bracis-28410 178 88 defined define VERB bracis-28410 178 89 in in ADP bracis-28410 178 90 eq eq PROPN bracis-28410 178 91 . . PUNCT bracis-28410 178 92     SPACE bracis-28410 178 93 3 3 NUM bracis-28410 178 94 at at ADP bracis-28410 178 95 m m PROPN bracis-28410 178 96 and and CCONJ bracis-28410 178 97 j j PROPN bracis-28410 178 98 scales scale NOUN bracis-28410 178 99 , , PUNCT bracis-28410 178 100 respectively respectively ADV bracis-28410 178 101 , , PUNCT bracis-28410 178 102 while while SCONJ bracis-28410 178 103 the the DET bracis-28410 178 104 \(\alpha \(\alpha NOUN bracis-28410 178 105 \ \ NOUN bracis-28410 178 106 ) ) PUNCT bracis-28410 178 107 and and CCONJ bracis-28410 178 108 \(\beta \(\beta ADP bracis-28410 178 109 _ _ PRON bracis-28410 178 110 { { PUNCT bracis-28410 178 111 j}\ j}\ NOUN bracis-28410 178 112 ) ) PUNCT bracis-28410 178 113 exponents exponent NOUN bracis-28410 178 114 are be AUX bracis-28410 178 115 used use VERB bracis-28410 178 116 to to PART bracis-28410 178 117 adjust adjust VERB bracis-28410 178 118 the the DET bracis-28410 178 119 relative relative ADJ bracis-28410 178 120 significance significance NOUN bracis-28410 178 121 of of ADP bracis-28410 178 122 different different ADJ bracis-28410 178 123 components component NOUN bracis-28410 178 124 . . PUNCT bracis-28410 179 1 it it PRON bracis-28410 179 2 is be AUX bracis-28410 179 3 worth worth ADJ bracis-28410 179 4 noting note VERB bracis-28410 179 5 that that SCONJ bracis-28410 179 6 the the DET bracis-28410 179 7 luminance luminance NOUN bracis-28410 179 8 comparison comparison NOUN bracis-28410 179 9 ( ( PUNCT bracis-28410 179 10 \(l_{m}\ \(l_{m}\ X bracis-28410 179 11 ) ) PUNCT bracis-28410 179 12 ) ) PUNCT bracis-28410 179 13 is be AUX bracis-28410 179 14 calculated calculate VERB bracis-28410 179 15 only only ADV bracis-28410 179 16 at at ADP bracis-28410 179 17 m m NOUN bracis-28410 179 18 scale scale NOUN bracis-28410 179 19 , , PUNCT bracis-28410 179 20 while while SCONJ bracis-28410 179 21 contrast contrast NOUN bracis-28410 179 22 and and CCONJ bracis-28410 179 23 structure structure NOUN bracis-28410 179 24 comparisons comparison NOUN bracis-28410 179 25 ( ( PUNCT bracis-28410 179 26 \(cs_{j}\ \(cs_{j}\ NUM bracis-28410 179 27 ) ) PUNCT bracis-28410 179 28 ) ) PUNCT bracis-28410 179 29 at at ADP bracis-28410 179 30 each each DET bracis-28410 179 31 scale scale NOUN bracis-28410 179 32 . . PUNCT bracis-28410 180 1 as as ADP bracis-28410 180 2 with with ADP bracis-28410 180 3 ssim ssim NOUN bracis-28410 180 4 , , PUNCT bracis-28410 180 5 the the DET bracis-28410 180 6 largest large ADJ bracis-28410 180 7 possible possible ADJ bracis-28410 180 8 value value NOUN bracis-28410 180 9 for for ADP bracis-28410 180 10 ms ms NOUN bracis-28410 180 11 - - PUNCT bracis-28410 180 12 ssim ssim NOUN bracis-28410 180 13 is be AUX bracis-28410 180 14 1 1 NUM bracis-28410 180 15 , , PUNCT bracis-28410 180 16 so so SCONJ bracis-28410 180 17 we we PRON bracis-28410 180 18 can can AUX bracis-28410 180 19 use use VERB bracis-28410 180 20 it it PRON bracis-28410 180 21 in in ADP bracis-28410 180 22 a a DET bracis-28410 180 23 loss loss NOUN bracis-28410 180 24 function function NOUN bracis-28410 180 25 in in ADP bracis-28410 180 26 the the DET bracis-28410 180 27 form form NOUN bracis-28410 180 28 : : PUNCT bracis-28410 180 29 $ $ SYM bracis-28410 180 30 $ $ SYM bracis-28410 180 31 \begin{aligned \begin{aligne VERB bracis-28410 180 32 } } PUNCT bracis-28410 180 33 \mathcal \mathcal ADJ bracis-28410 180 34 { { PUNCT bracis-28410 180 35 l}^{\text l}^{\text X bracis-28410 180 36 { { PUNCT bracis-28410 180 37 ms ms PROPN bracis-28410 180 38 - - PROPN bracis-28410 180 39 ssim}}(p)=1-\text ssim}}(p)=1-\text PROPN bracis-28410 180 40 { { PUNCT bracis-28410 180 41 ms ms PROPN bracis-28410 180 42 - - PUNCT bracis-28410 180 43 ssim}(\tilde{p ssim}(\tilde{p NOUN bracis-28410 180 44 } } PUNCT bracis-28410 180 45 ) ) PUNCT bracis-28410 180 46 . . PUNCT bracis-28410 181 1 \end{aligned}$$ \end{aligned}$$ PROPN bracis-28410 181 2 ( ( PUNCT bracis-28410 181 3 4 4 X bracis-28410 181 4 ) ) PUNCT bracis-28410 181 5 we we PRON bracis-28410 181 6 also also ADV bracis-28410 181 7 explored explore VERB bracis-28410 181 8 the the DET bracis-28410 181 9 combination combination NOUN bracis-28410 181 10 of of ADP bracis-28410 181 11 functions function NOUN bracis-28410 181 12 employing employ VERB bracis-28410 181 13 several several ADJ bracis-28410 181 14 different different ADJ bracis-28410 181 15 layers layer NOUN bracis-28410 181 16 of of ADP bracis-28410 181 17 the the DET bracis-28410 181 18 network network NOUN bracis-28410 181 19 , , PUNCT bracis-28410 181 20 as as SCONJ bracis-28410 181 21 suggested suggest VERB bracis-28410 181 22 by by ADP bracis-28410 181 23 xu xu PROPN bracis-28410 181 24 et et PROPN bracis-28410 181 25 al al PROPN bracis-28410 181 26 . . PUNCT bracis-28410 181 27     SPACE bracis-28410 182 1 [ [ X bracis-28410 182 2 58 58 NUM bracis-28410 182 3 ] ] PUNCT bracis-28410 182 4 , , PUNCT bracis-28410 182 5 in in ADP bracis-28410 182 6 which which PRON bracis-28410 182 7 the the DET bracis-28410 182 8 weighted weighted ADJ bracis-28410 182 9 sum sum NOUN bracis-28410 182 10 between between ADP bracis-28410 182 11 the the DET bracis-28410 182 12 calculation calculation NOUN bracis-28410 182 13 of of ADP bracis-28410 182 14 the the DET bracis-28410 182 15 l1 l1 PROPN bracis-28410 182 16 function function NOUN bracis-28410 182 17 after after ADP bracis-28410 182 18 two two NUM bracis-28410 182 19 , , PUNCT bracis-28410 182 20 three three NUM bracis-28410 182 21 and and CCONJ bracis-28410 182 22 four four NUM bracis-28410 182 23 residual residual ADJ bracis-28410 182 24 blocks block NOUN bracis-28410 182 25 are be AUX bracis-28410 182 26 used use VERB bracis-28410 182 27 , , PUNCT bracis-28410 182 28 with with ADP bracis-28410 182 29 weights weight NOUN bracis-28410 182 30 of of ADP bracis-28410 182 31 0.5 0.5 NUM bracis-28410 182 32 , , PUNCT bracis-28410 182 33 0.5 0.5 NUM bracis-28410 182 34 and and CCONJ bracis-28410 182 35 1 1 NUM bracis-28410 182 36 , , PUNCT bracis-28410 182 37 respectively respectively ADV bracis-28410 182 38 . . PUNCT bracis-28410 183 1 after after ADP bracis-28410 183 2 each each DET bracis-28410 183 3 residual residual ADJ bracis-28410 183 4 block block NOUN bracis-28410 183 5 composing compose VERB bracis-28410 183 6 the the DET bracis-28410 183 7 loss loss NOUN bracis-28410 183 8 , , PUNCT bracis-28410 183 9 we we PRON bracis-28410 183 10 add add VERB bracis-28410 183 11 a a DET bracis-28410 183 12 network network NOUN bracis-28410 183 13 based base VERB bracis-28410 183 14 on on ADP bracis-28410 183 15 subpixel subpixel ADJ bracis-28410 183 16 convolutions convolution NOUN bracis-28410 183 17 to to PART bracis-28410 183 18 perform perform VERB bracis-28410 183 19 upscaling upscale VERB bracis-28410 183 20 . . PUNCT bracis-28410 184 1 besides besides SCONJ bracis-28410 184 2 the the DET bracis-28410 184 3 above above ADJ bracis-28410 184 4 settings setting NOUN bracis-28410 184 5 , , PUNCT bracis-28410 184 6 we we PRON bracis-28410 184 7 investigate investigate VERB bracis-28410 184 8 the the DET bracis-28410 184 9 benefits benefit NOUN bracis-28410 184 10 of of ADP bracis-28410 184 11 edge edge NOUN bracis-28410 184 12 maps map NOUN bracis-28410 184 13 in in ADP bracis-28410 184 14 the the DET bracis-28410 184 15 loss loss NOUN bracis-28410 184 16 function function NOUN bracis-28410 184 17 , , PUNCT bracis-28410 184 18 since since SCONJ bracis-28410 184 19 l1 l1 PROPN bracis-28410 184 20 may may AUX bracis-28410 184 21 smooth smooth VERB bracis-28410 184 22 the the DET bracis-28410 184 23 edges edge NOUN bracis-28410 184 24 . . PUNCT bracis-28410 185 1 similarly similarly ADV bracis-28410 185 2 to to ADP bracis-28410 185 3 pandey pandey PROPN bracis-28410 185 4 et et PROPN bracis-28410 185 5 al al PROPN bracis-28410 185 6 . . PUNCT bracis-28410 185 7     SPACE bracis-28410 186 1 [ [ X bracis-28410 186 2 39 39 NUM bracis-28410 186 3 ] ] PUNCT bracis-28410 186 4 , , PUNCT bracis-28410 186 5 we we PRON bracis-28410 186 6 evaluated evaluate VERB bracis-28410 186 7 a a DET bracis-28410 186 8 combination combination NOUN bracis-28410 186 9 of of ADP bracis-28410 186 10 the the DET bracis-28410 186 11 l1 l1 PROPN bracis-28410 186 12 using use VERB bracis-28410 186 13 the the DET bracis-28410 186 14 sr sr PROPN bracis-28410 186 15 and and CCONJ bracis-28410 186 16 hr hr NOUN bracis-28410 186 17 images image NOUN bracis-28410 186 18 and and CCONJ bracis-28410 186 19 the the DET bracis-28410 186 20 l1 l1 PROPN bracis-28410 186 21 between between ADP bracis-28410 186 22 its its PRON bracis-28410 186 23 edge edge NOUN bracis-28410 186 24 maps map NOUN bracis-28410 186 25 . . PUNCT bracis-28410 187 1 however however ADV bracis-28410 187 2 , , PUNCT bracis-28410 187 3 although although SCONJ bracis-28410 187 4 their their PRON bracis-28410 187 5 work work NOUN bracis-28410 187 6 uses use VERB bracis-28410 187 7 the the DET bracis-28410 187 8 canny canny ADJ bracis-28410 187 9 operator operator NOUN bracis-28410 187 10     SPACE bracis-28410 187 11 [ [ X bracis-28410 187 12 9 9 NUM bracis-28410 187 13 ] ] PUNCT bracis-28410 187 14 to to PART bracis-28410 187 15 generate generate VERB bracis-28410 187 16 the the DET bracis-28410 187 17 edge edge NOUN bracis-28410 187 18 map map NOUN bracis-28410 187 19 , , PUNCT bracis-28410 187 20 our our PRON bracis-28410 187 21 work work NOUN bracis-28410 187 22 investigated investigate VERB bracis-28410 187 23 the the DET bracis-28410 187 24 usage usage NOUN bracis-28410 187 25 of of ADP bracis-28410 187 26 the the DET bracis-28410 187 27 sobel sobel NOUN bracis-28410 187 28 operator operator NOUN bracis-28410 187 29     SPACE bracis-28410 188 1 [ [ X bracis-28410 188 2 47 47 NUM bracis-28410 188 3 ] ] PUNCT bracis-28410 188 4 . . PUNCT bracis-28410 189 1 we we PRON bracis-28410 189 2 also also ADV bracis-28410 189 3 consider consider VERB bracis-28410 189 4 the the DET bracis-28410 189 5 three three NUM bracis-28410 189 6 - - PUNCT bracis-28410 189 7 component component NOUN bracis-28410 189 8 weighted weight VERB bracis-28410 189 9 psnr psnr NOUN bracis-28410 189 10 ( ( PUNCT bracis-28410 189 11 3 3 NUM bracis-28410 189 12 - - PUNCT bracis-28410 189 13 psnr psnr NOUN bracis-28410 189 14 ) ) PUNCT bracis-28410 189 15 and and CCONJ bracis-28410 189 16 ssim ssim NOUN bracis-28410 189 17 ( ( PUNCT bracis-28410 189 18 3 3 NUM bracis-28410 189 19 - - PUNCT bracis-28410 189 20 ssim ssim NOUN bracis-28410 189 21 ) ) PUNCT bracis-28410 189 22 loss loss NOUN bracis-28410 189 23 functions function NOUN bracis-28410 189 24     SPACE bracis-28410 190 1 [ [ X bracis-28410 190 2 31 31 NUM bracis-28410 190 3 ] ] PUNCT bracis-28410 190 4 . . PUNCT bracis-28410 191 1 importantly importantly ADV bracis-28410 191 2 , , PUNCT bracis-28410 191 3 such such ADJ bracis-28410 191 4 metrics metric NOUN bracis-28410 191 5 can can AUX bracis-28410 191 6 measure measure VERB bracis-28410 191 7 the the DET bracis-28410 191 8 quality quality NOUN bracis-28410 191 9 of of ADP bracis-28410 191 10 images image NOUN bracis-28410 191 11 and and CCONJ bracis-28410 191 12 videos video NOUN bracis-28410 191 13 . . PUNCT bracis-28410 192 1 this this DET bracis-28410 192 2 approach approach NOUN bracis-28410 192 3 breaks break VERB bracis-28410 192 4 up up ADP bracis-28410 192 5 an an DET bracis-28410 192 6 image image NOUN bracis-28410 192 7 into into ADP bracis-28410 192 8 three three NUM bracis-28410 192 9 parts part NOUN bracis-28410 192 10 : : PUNCT bracis-28410 192 11 edges edge NOUN bracis-28410 192 12 , , PUNCT bracis-28410 192 13 textures texture NOUN bracis-28410 192 14 , , PUNCT bracis-28410 192 15 and and CCONJ bracis-28410 192 16 more more ADV bracis-28410 192 17 homogeneous homogeneous ADJ bracis-28410 192 18 regions region NOUN bracis-28410 192 19 . . PUNCT bracis-28410 193 1 to to PART bracis-28410 193 2 do do VERB bracis-28410 193 3 this this PRON bracis-28410 193 4 , , PUNCT bracis-28410 193 5 the the DET bracis-28410 193 6 sobel sobel NOUN bracis-28410 193 7 operator operator NOUN bracis-28410 193 8 is be AUX bracis-28410 193 9 applied apply VERB bracis-28410 193 10 to to ADP bracis-28410 193 11 the the DET bracis-28410 193 12 luminance luminance NOUN bracis-28410 193 13 channel channel NOUN bracis-28410 193 14 of of ADP bracis-28410 193 15 the the DET bracis-28410 193 16 image image NOUN bracis-28410 193 17 and and CCONJ bracis-28410 193 18 , , PUNCT bracis-28410 193 19 from from ADP bracis-28410 193 20 the the DET bracis-28410 193 21 highest high ADJ bracis-28410 193 22 calculated calculate VERB bracis-28410 193 23 value value NOUN bracis-28410 193 24 and and CCONJ bracis-28410 193 25 some some DET bracis-28410 193 26 pre pre ADJ bracis-28410 193 27 - - ADJ bracis-28410 193 28 established established ADJ bracis-28410 193 29 values value NOUN bracis-28410 193 30 , , PUNCT bracis-28410 193 31 the the DET bracis-28410 193 32 thresholds threshold NOUN bracis-28410 193 33 that that PRON bracis-28410 193 34 delimit delimit VERB bracis-28410 193 35 each each DET bracis-28410 193 36 region region NOUN bracis-28410 193 37 are be AUX bracis-28410 193 38 calculated calculate VERB bracis-28410 193 39 . . PUNCT bracis-28410 194 1 the the DET bracis-28410 194 2 value value NOUN bracis-28410 194 3 of of ADP bracis-28410 194 4 each each DET bracis-28410 194 5 metric metric NOUN bracis-28410 194 6 is be AUX bracis-28410 194 7 calculated calculate VERB bracis-28410 194 8 by by ADP bracis-28410 194 9 applying apply VERB bracis-28410 194 10 different different ADJ bracis-28410 194 11 weights weight NOUN bracis-28410 194 12 for for ADP bracis-28410 194 13 each each DET bracis-28410 194 14 region region NOUN bracis-28410 194 15 . . PUNCT bracis-28410 195 1 li li PROPN bracis-28410 195 2 and and CCONJ bracis-28410 195 3 bovik bovik NOUN bracis-28410 195 4     SPACE bracis-28410 196 1 [ [ X bracis-28410 196 2 31 31 NUM bracis-28410 196 3 ] ] PUNCT bracis-28410 196 4 showed show VERB bracis-28410 196 5 that that SCONJ bracis-28410 196 6 the the DET bracis-28410 196 7 weights weight NOUN bracis-28410 196 8 that that PRON bracis-28410 196 9 achieved achieve VERB bracis-28410 196 10 the the DET bracis-28410 196 11 best good ADJ bracis-28410 196 12 results result NOUN bracis-28410 196 13 were be AUX bracis-28410 196 14 0.7 0.7 NUM bracis-28410 196 15 , , PUNCT bracis-28410 196 16 0.15 0.15 NUM bracis-28410 196 17 and and CCONJ bracis-28410 196 18 0.15 0.15 NUM bracis-28410 196 19 for for ADP bracis-28410 196 20 3 3 NUM bracis-28410 196 21 - - PUNCT bracis-28410 196 22 psnr psnr NOUN bracis-28410 196 23 , , PUNCT bracis-28410 196 24 and and CCONJ bracis-28410 196 25 1 1 NUM bracis-28410 196 26 , , PUNCT bracis-28410 196 27 0 0 NUM bracis-28410 196 28 and and CCONJ bracis-28410 196 29 0 0 NUM bracis-28410 196 30 for for ADP bracis-28410 196 31 3 3 NUM bracis-28410 196 32 - - PUNCT bracis-28410 196 33 ssim ssim NOUN bracis-28410 196 34 , , PUNCT bracis-28410 196 35 considering consider VERB bracis-28410 196 36 edges edge NOUN bracis-28410 196 37 , , PUNCT bracis-28410 196 38 textures texture NOUN bracis-28410 196 39 , , PUNCT bracis-28410 196 40 and and CCONJ bracis-28410 196 41 homogeneous homogeneous ADJ bracis-28410 196 42 regions region NOUN bracis-28410 196 43 , , PUNCT bracis-28410 196 44 respectively respectively ADV bracis-28410 196 45 . . PUNCT bracis-28410 197 1 these these DET bracis-28410 197 2 values value NOUN bracis-28410 197 3 are be AUX bracis-28410 197 4 consistent consistent ADJ bracis-28410 197 5 with with ADP bracis-28410 197 6 the the DET bracis-28410 197 7 observation observation NOUN bracis-28410 197 8 that that SCONJ bracis-28410 197 9 perturbations perturbation NOUN bracis-28410 197 10 at at ADP bracis-28410 197 11 the the DET bracis-28410 197 12 edges edge NOUN bracis-28410 197 13 of of ADP bracis-28410 197 14 an an DET bracis-28410 197 15 object object NOUN bracis-28410 197 16 are be AUX bracis-28410 197 17 perceptually perceptually ADV bracis-28410 197 18 more more ADV bracis-28410 197 19 significant significant ADJ bracis-28410 197 20 than than ADP bracis-28410 197 21 in in ADP bracis-28410 197 22 other other ADJ bracis-28410 197 23 areas area NOUN bracis-28410 197 24 . . PUNCT bracis-28410 198 1 based base VERB bracis-28410 198 2 on on ADP bracis-28410 198 3 recent recent ADJ bracis-28410 198 4 solutions solution NOUN bracis-28410 198 5 available available ADJ bracis-28410 198 6 in in ADP bracis-28410 198 7 the the DET bracis-28410 198 8 literature literature NOUN bracis-28410 198 9 , , PUNCT bracis-28410 198 10 barron barron PROPN bracis-28410 198 11     SPACE bracis-28410 198 12 [ [ X bracis-28410 198 13 5 5 NUM bracis-28410 198 14 ] ] PUNCT bracis-28410 198 15 presented present VERB bracis-28410 198 16 a a DET bracis-28410 198 17 loss loss NOUN bracis-28410 198 18 function function NOUN bracis-28410 198 19 that that PRON bracis-28410 198 20 is be AUX bracis-28410 198 21 a a DET bracis-28410 198 22 superset superset NOUN bracis-28410 198 23 of of ADP bracis-28410 198 24 cauchy cauchy PROPN bracis-28410 198 25 / / SYM bracis-28410 198 26 lorentzian lorentzian ADJ bracis-28410 198 27 , , PUNCT bracis-28410 198 28 geman geman NOUN bracis-28410 198 29 - - PUNCT bracis-28410 198 30 mcclure mcclure PROPN bracis-28410 198 31 , , PUNCT bracis-28410 198 32 welsch welsch PROPN bracis-28410 198 33 / / SYM bracis-28410 198 34 leclerc leclerc PROPN bracis-28410 198 35 , , PUNCT bracis-28410 198 36 generalized generalized ADJ bracis-28410 198 37 charbonnier charbonnier NOUN bracis-28410 198 38 , , PUNCT bracis-28410 198 39 charbonnier charbonnier NOUN bracis-28410 198 40 / / SYM bracis-28410 198 41 pseudo pseudo NOUN bracis-28410 198 42 - - NOUN bracis-28410 198 43 huber huber NOUN bracis-28410 198 44 / / SYM bracis-28410 198 45 l1 l1 NOUN bracis-28410 198 46 - - PUNCT bracis-28410 198 47 l2 l2 NOUN bracis-28410 198 48 , , PUNCT bracis-28410 198 49 and and CCONJ bracis-28410 198 50 l2 l2 NOUN bracis-28410 198 51 . . PUNCT bracis-28410 199 1 this this DET bracis-28410 199 2 function function NOUN bracis-28410 199 3 has have VERB bracis-28410 199 4 two two NUM bracis-28410 199 5 hyperparameters hyperparameter NOUN bracis-28410 199 6 : : PUNCT bracis-28410 199 7 robustness robustness NOUN bracis-28410 199 8 ( ( PUNCT bracis-28410 199 9 \(\alpha \(\alpha PROPN bracis-28410 199 10 \ \ NOUN bracis-28410 199 11 ) ) PUNCT bracis-28410 199 12 ) ) PUNCT bracis-28410 199 13 and and CCONJ bracis-28410 199 14 scale scale NOUN bracis-28410 199 15 ( ( PUNCT bracis-28410 199 16 c c NOUN bracis-28410 199 17 ) ) PUNCT bracis-28410 199 18 , , PUNCT bracis-28410 199 19 with with ADP bracis-28410 199 20 the the DET bracis-28410 199 21 variation variation NOUN bracis-28410 199 22 of of ADP bracis-28410 199 23 which which PRON bracis-28410 199 24 is be AUX bracis-28410 199 25 possible possible ADJ bracis-28410 199 26 to to PART bracis-28410 199 27 reach reach VERB bracis-28410 199 28 all all DET bracis-28410 199 29 previous previous ADJ bracis-28410 199 30 functions function NOUN bracis-28410 199 31 as as ADP bracis-28410 199 32 specific specific ADJ bracis-28410 199 33 cases case NOUN bracis-28410 199 34 . . PUNCT bracis-28410 200 1 the the DET bracis-28410 200 2 general general ADJ bracis-28410 200 3 loss loss NOUN bracis-28410 200 4 function function NOUN bracis-28410 200 5 is be AUX bracis-28410 200 6 calculated calculate VERB bracis-28410 200 7 as as SCONJ bracis-28410 200 8 follows follow VERB bracis-28410 200 9 : : PUNCT bracis-28410 200 10 $ $ SYM bracis-28410 200 11 $ $ SYM bracis-28410 200 12 \begin{aligned \begin{aligne VERB bracis-28410 200 13 } } PUNCT bracis-28410 200 14 \mathcal \mathcal ADJ bracis-28410 200 15 { { PUNCT bracis-28410 200 16 l}^{\text l}^{\text ADJ bracis-28410 200 17 { { PUNCT bracis-28410 200 18 general}}(x general}}(x PROPN bracis-28410 200 19 , , PUNCT bracis-28410 200 20 \alpha \alpha ADV bracis-28410 200 21 , , PUNCT bracis-28410 200 22 c c X bracis-28410 200 23 ) ) PUNCT bracis-28410 200 24 \!=\ \!=\ NOUN bracis-28410 200 25 ! ! PUNCT bracis-28410 201 1 \left\ \left\ PROPN bracis-28410 201 2 { { PUNCT bracis-28410 201 3 \begin{array}{ll}{\frac{1}{2}(x \begin{array}{ll}{\frac{1}{2}(x PROPN bracis-28410 201 4 / / SYM bracis-28410 201 5 c)^{2 c)^{2 NOUN bracis-28410 201 6 } } PUNCT bracis-28410 201 7 } } PUNCT bracis-28410 201 8 \!\!&{}\!\ \!\!&{}\!\ NOUN bracis-28410 201 9 ! ! PUNCT bracis-28410 202 1 { { PUNCT bracis-28410 202 2 \textrm \textrm ADJ bracis-28410 202 3 { { PUNCT bracis-28410 202 4 if if SCONJ bracis-28410 202 5 } } PUNCT bracis-28410 202 6 \ \ NOUN bracis-28410 202 7 \alpha \alpha ADP bracis-28410 202 8 = = NOUN bracis-28410 202 9 2 2 NUM bracis-28410 202 10 } } PUNCT bracis-28410 202 11 \\ \\ NOUN bracis-28410 202 12 { { PUNCT bracis-28410 202 13 \log \log PROPN bracis-28410 202 14 \left \left PROPN bracis-28410 202 15 ( ( PUNCT bracis-28410 202 16 \frac{1}{2}(x \frac{1}{2}(x NOUN bracis-28410 202 17 / / SYM bracis-28410 202 18 c)^{2}+1\right c)^{2}+1\right NOUN bracis-28410 202 19 ) ) PUNCT bracis-28410 202 20 } } PUNCT bracis-28410 202 21 \!\!&{}\!\ \!\!&{}\!\ NOUN bracis-28410 202 22 ! ! PUNCT bracis-28410 203 1 { { PUNCT bracis-28410 203 2 \textrm \textrm ADJ bracis-28410 203 3 { { PUNCT bracis-28410 203 4 if if SCONJ bracis-28410 203 5 } } PUNCT bracis-28410 203 6 \ \ NOUN bracis-28410 203 7 \alpha \alpha ADP bracis-28410 203 8 = = NOUN bracis-28410 203 9 0 0 NUM bracis-28410 203 10 } } PUNCT bracis-28410 203 11 \\ \\ NOUN bracis-28410 203 12 { { PUNCT bracis-28410 203 13 1-\exp 1-\exp NUM bracis-28410 203 14 \left \left PROPN bracis-28410 203 15 ( ( PUNCT bracis-28410 203 16 -\frac{1}{2}(x -\frac{1}{2}(x ADJ bracis-28410 203 17 / / SYM bracis-28410 203 18 c)^{2}\right c)^{2}\right NUM bracis-28410 203 19 ) ) PUNCT bracis-28410 203 20 } } PUNCT bracis-28410 203 21 \!\!&{}\!\ \!\!&{}\!\ NOUN bracis-28410 203 22 ! ! PUNCT bracis-28410 204 1 { { PUNCT bracis-28410 204 2 \textrm \textrm ADJ bracis-28410 204 3 { { PUNCT bracis-28410 204 4 if if SCONJ bracis-28410 204 5 } } PUNCT bracis-28410 204 6 \ \ NOUN bracis-28410 204 7 \alpha \alpha SCONJ bracis-28410 204 8 = = VERB bracis-28410 204 9 -\infty -\infty ADJ bracis-28410 204 10 } } PUNCT bracis-28410 204 11 \\ \\ NOUN bracis-28410 204 12 { { PUNCT bracis-28410 204 13 \frac{|2-\alpha \frac{|2-\alpha PROPN bracis-28410 204 14 |}{\alpha |}{\alpha PROPN bracis-28410 204 15 } } PUNCT bracis-28410 204 16 \left \left PROPN bracis-28410 204 17 ( ( PUNCT bracis-28410 204 18 \left \left PROPN bracis-28410 204 19 ( ( PUNCT bracis-28410 204 20 \frac{(x \frac{(x PROPN bracis-28410 204 21 / / SYM bracis-28410 204 22 c)^{2}}{|2-\alpha c)^{2}}{|2-\alpha PROPN bracis-28410 204 23 |}+1\right |}+1\right NOUN bracis-28410 204 24 ) ) PUNCT bracis-28410 204 25 ^{(\alpha ^{(\alpha PUNCT bracis-28410 204 26 / / SYM bracis-28410 204 27 2)}-1\right 2)}-1\right NUM bracis-28410 204 28 ) ) PUNCT bracis-28410 204 29 } } PUNCT bracis-28410 204 30 \!\!&{}\!\ \!\!&{}\!\ NOUN bracis-28410 204 31 ! ! PUNCT bracis-28410 205 1 { { PUNCT bracis-28410 205 2 \textrm \textrm ADJ bracis-28410 205 3 { { PUNCT bracis-28410 205 4 otherwise otherwise ADV bracis-28410 205 5 } } PUNCT bracis-28410 205 6 } } PUNCT bracis-28410 205 7 \end{array}\right \end{array}\right PROPN bracis-28410 205 8 . . PUNCT bracis-28410 206 1 \end{aligned}$$ \end{aligned}$$ PROPN bracis-28410 206 2 where where SCONJ bracis-28410 206 3 x x PRON bracis-28410 206 4 is be AUX bracis-28410 206 5 the the DET bracis-28410 206 6 difference difference NOUN bracis-28410 206 7 between between ADP bracis-28410 206 8 the the DET bracis-28410 206 9 hr hr NOUN bracis-28410 206 10 and and CCONJ bracis-28410 206 11 sr sr PROPN bracis-28410 206 12 pixel pixel PROPN bracis-28410 206 13 values value NOUN bracis-28410 206 14 . . PUNCT bracis-28410 207 1 barron barron PROPN bracis-28410 207 2 also also ADV bracis-28410 207 3 showed show VERB bracis-28410 207 4 that that SCONJ bracis-28410 207 5 it it PRON bracis-28410 207 6 is be AUX bracis-28410 207 7 possible possible ADJ bracis-28410 207 8 to to PART bracis-28410 207 9 modify modify VERB bracis-28410 207 10 its its PRON bracis-28410 207 11 function function NOUN bracis-28410 207 12 so so SCONJ bracis-28410 207 13 that that SCONJ bracis-28410 207 14 the the DET bracis-28410 207 15 network network NOUN bracis-28410 207 16 learns learn VERB bracis-28410 207 17 optimal optimal ADJ bracis-28410 207 18 values value NOUN bracis-28410 207 19 for for ADP bracis-28410 207 20 the the DET bracis-28410 207 21 \(\alpha \(\alpha NOUN bracis-28410 207 22 \ \ NOUN bracis-28410 207 23 ) ) PUNCT bracis-28410 207 24 and and CCONJ bracis-28410 207 25 c c NOUN bracis-28410 207 26 parameters parameter NOUN bracis-28410 207 27 , , PUNCT bracis-28410 207 28 thus thus ADV bracis-28410 207 29 providing provide VERB bracis-28410 207 30 an an DET bracis-28410 207 31 appropriate appropriate ADJ bracis-28410 207 32 exploration exploration NOUN bracis-28410 207 33 by by ADP bracis-28410 207 34 the the DET bracis-28410 207 35 network network NOUN bracis-28410 207 36 of of ADP bracis-28410 207 37 different different ADJ bracis-28410 207 38 error error NOUN bracis-28410 207 39 functions function NOUN bracis-28410 207 40 . . PUNCT bracis-28410 208 1 due due ADP bracis-28410 208 2 to to ADP bracis-28410 208 3 its its PRON bracis-28410 208 4 unique unique ADJ bracis-28410 208 5 features feature NOUN bracis-28410 208 6 , , PUNCT bracis-28410 208 7 we we PRON bracis-28410 208 8 use use VERB bracis-28410 208 9 the the DET bracis-28410 208 10 adaptive adaptive ADJ bracis-28410 208 11 loss loss NOUN bracis-28410 208 12 function function NOUN bracis-28410 208 13 for for ADP bracis-28410 208 14 the the DET bracis-28410 208 15 srcaps srcap NOUN bracis-28410 208 16 training training NOUN bracis-28410 208 17 . . PUNCT bracis-28410 209 1 4 4 NUM bracis-28410 209 2 experimental experimental ADJ bracis-28410 209 3 results result NOUN bracis-28410 209 4 datasets dataset NOUN bracis-28410 209 5 . . PUNCT bracis-28410 210 1 in in ADP bracis-28410 210 2 this this DET bracis-28410 210 3 work work NOUN bracis-28410 210 4 , , PUNCT bracis-28410 210 5 we we PRON bracis-28410 210 6 employed employ VERB bracis-28410 210 7 datasets dataset NOUN bracis-28410 210 8 widely widely ADV bracis-28410 210 9 used use VERB bracis-28410 210 10 in in ADP bracis-28410 210 11 the the DET bracis-28410 210 12 literature literature NOUN bracis-28410 210 13     SPACE bracis-28410 211 1 [ [ X bracis-28410 211 2 4 4 NUM bracis-28410 211 3 , , PUNCT bracis-28410 211 4 56 56 NUM bracis-28410 211 5 ] ] PUNCT bracis-28410 211 6 . . PUNCT bracis-28410 212 1 currently currently ADV bracis-28410 212 2 , , PUNCT bracis-28410 212 3 the the DET bracis-28410 212 4 div2k div2k PROPN bracis-28410 212 5 training training NOUN bracis-28410 212 6 set set NOUN bracis-28410 212 7 is be AUX bracis-28410 212 8 used use VERB bracis-28410 212 9 in in ADP bracis-28410 212 10 the the DET bracis-28410 212 11 training training NOUN bracis-28410 212 12 of of ADP bracis-28410 212 13 neural neural ADJ bracis-28410 212 14 networks network NOUN bracis-28410 212 15 for for ADP bracis-28410 212 16 the the DET bracis-28410 212 17 super super ADJ bracis-28410 212 18 - - ADJ bracis-28410 212 19 resolution resolution ADJ bracis-28410 212 20 problem problem NOUN bracis-28410 212 21 , , PUNCT bracis-28410 212 22 while while SCONJ bracis-28410 212 23 the the DET bracis-28410 212 24 validation validation NOUN bracis-28410 212 25 set set NOUN bracis-28410 212 26 of of ADP bracis-28410 212 27 div2k div2k PROPN bracis-28410 212 28 , , PUNCT bracis-28410 212 29 b100 b100 PROPN bracis-28410 212 30 , , PUNCT bracis-28410 212 31 set5 set5 PROPN bracis-28410 212 32 , , PUNCT bracis-28410 212 33 set14 set14 NOUN bracis-28410 212 34 , , PUNCT bracis-28410 212 35 and and CCONJ bracis-28410 212 36 urban100 urban100 PROPN bracis-28410 212 37 are be AUX bracis-28410 212 38 used use VERB bracis-28410 212 39 to to PART bracis-28410 212 40 validate validate VERB bracis-28410 212 41 the the DET bracis-28410 212 42 results result NOUN bracis-28410 212 43 . . PUNCT bracis-28410 213 1 all all DET bracis-28410 213 2 datasets dataset NOUN bracis-28410 213 3 are be AUX bracis-28410 213 4 composed compose VERB bracis-28410 213 5 of of ADP bracis-28410 213 6 original original ADJ bracis-28410 213 7 versions version NOUN bracis-28410 213 8 of of ADP bracis-28410 213 9 the the DET bracis-28410 213 10 images image NOUN bracis-28410 213 11 ( ( PUNCT bracis-28410 213 12 hr hr NOUN bracis-28410 213 13 ) ) PUNCT bracis-28410 213 14 and and CCONJ bracis-28410 213 15 their their PRON bracis-28410 213 16 reduced reduce VERB bracis-28410 213 17 versions version NOUN bracis-28410 213 18 by by ADP bracis-28410 213 19 bicubic bicubic ADJ bracis-28410 213 20 interpolation interpolation NOUN bracis-28410 213 21 algorithm algorithm NOUN bracis-28410 213 22 for for ADP bracis-28410 213 23 \(2\times \(2\time NOUN bracis-28410 213 24 \ \ NOUN bracis-28410 213 25 ) ) PUNCT bracis-28410 213 26 , , PUNCT bracis-28410 213 27 \(3\times \(3\time VERB bracis-28410 213 28 \ \ NOUN bracis-28410 213 29 ) ) PUNCT bracis-28410 213 30 , , PUNCT bracis-28410 213 31 and and CCONJ bracis-28410 213 32 \(4\times \(4\times ADP bracis-28410 213 33 \ \ NOUN bracis-28410 213 34 ) ) PUNCT bracis-28410 213 35 . . PUNCT bracis-28410 214 1 in in ADP bracis-28410 214 2 this this DET bracis-28410 214 3 work work NOUN bracis-28410 214 4 , , PUNCT bracis-28410 214 5 we we PRON bracis-28410 214 6 focus focus VERB bracis-28410 214 7 on on ADP bracis-28410 214 8 the the DET bracis-28410 214 9 \(4\times \(4\times PROPN bracis-28410 214 10 \ \ NOUN bracis-28410 214 11 ) ) PUNCT bracis-28410 214 12 scale scale NOUN bracis-28410 214 13 factor factor NOUN bracis-28410 214 14 . . PUNCT bracis-28410 215 1 metrics metric NOUN bracis-28410 215 2 . . PUNCT bracis-28410 216 1 for for ADP bracis-28410 216 2 the the DET bracis-28410 216 3 validation validation NOUN bracis-28410 216 4 process process NOUN bracis-28410 216 5 of of ADP bracis-28410 216 6 the the DET bracis-28410 216 7 results result NOUN bracis-28410 216 8 obtained obtain VERB bracis-28410 216 9 , , PUNCT bracis-28410 216 10 we we PRON bracis-28410 216 11 employed employ VERB bracis-28410 216 12 metrics metric NOUN bracis-28410 216 13 commonly commonly ADV bracis-28410 216 14 used use VERB bracis-28410 216 15 in in ADP bracis-28410 216 16 the the DET bracis-28410 216 17 literature literature NOUN bracis-28410 216 18 . . PUNCT bracis-28410 217 1 more more ADV bracis-28410 217 2 specifically specifically ADV bracis-28410 217 3 , , PUNCT bracis-28410 217 4 peak peak NOUN bracis-28410 217 5 signal signal NOUN bracis-28410 217 6 - - PUNCT bracis-28410 217 7 to to ADP bracis-28410 217 8 - - PUNCT bracis-28410 217 9 noise noise NOUN bracis-28410 217 10 ratio ratio NOUN bracis-28410 217 11 ( ( PUNCT bracis-28410 217 12 psnr psnr NOUN bracis-28410 217 13 ) ) PUNCT bracis-28410 217 14     SPACE bracis-28410 218 1 [ [ X bracis-28410 218 2 17 17 NUM bracis-28410 218 3 ] ] PUNCT bracis-28410 218 4 , , PUNCT bracis-28410 218 5 structural structural ADJ bracis-28410 218 6 similarity similarity NOUN bracis-28410 218 7 index index NOUN bracis-28410 218 8 ( ( PUNCT bracis-28410 218 9 ssim ssim NOUN bracis-28410 218 10 ) ) PUNCT bracis-28410 218 11     SPACE bracis-28410 219 1 [ [ X bracis-28410 219 2 57 57 NUM bracis-28410 219 3 ] ] PUNCT bracis-28410 219 4 and and CCONJ bracis-28410 219 5 multi multi ADJ bracis-28410 219 6 - - ADJ bracis-28410 219 7 scale scale ADJ bracis-28410 219 8 structural structural ADJ bracis-28410 219 9 similarity similarity NOUN bracis-28410 219 10 index index NOUN bracis-28410 219 11 ( ( PUNCT bracis-28410 219 12 ms ms NOUN bracis-28410 219 13 - - PUNCT bracis-28410 219 14 ssim ssim NOUN bracis-28410 219 15 ) ) PUNCT bracis-28410 219 16     SPACE bracis-28410 220 1 [ [ X bracis-28410 220 2 55 55 NUM bracis-28410 220 3 ] ] PUNCT bracis-28410 220 4 . . PUNCT bracis-28410 221 1 due due ADP bracis-28410 221 2 to to ADP bracis-28410 221 3 space space NOUN bracis-28410 221 4 limitations limitation NOUN bracis-28410 221 5 , , PUNCT bracis-28410 221 6 we we PRON bracis-28410 221 7 refer refer VERB bracis-28410 221 8 interested interested ADJ bracis-28410 221 9 readers reader NOUN bracis-28410 221 10 to to ADP bracis-28410 221 11 the the DET bracis-28410 221 12 work work NOUN bracis-28410 221 13 developed develop VERB bracis-28410 221 14 by by ADP bracis-28410 221 15 wang wang PROPN bracis-28410 221 16 et et PROPN bracis-28410 221 17 al al PROPN bracis-28410 221 18 . . PUNCT bracis-28410 221 19     SPACE bracis-28410 222 1 [ [ X bracis-28410 222 2 56 56 NUM bracis-28410 222 3 ] ] PUNCT bracis-28410 222 4 for for ADP bracis-28410 222 5 a a DET bracis-28410 222 6 detailed detailed ADJ bracis-28410 222 7 formulation formulation NOUN bracis-28410 222 8 . . PUNCT bracis-28410 223 1 algorithms algorithm NOUN bracis-28410 223 2 that that PRON bracis-28410 223 3 measure measure VERB bracis-28410 223 4 the the DET bracis-28410 223 5 differences difference NOUN bracis-28410 223 6 between between ADP bracis-28410 223 7 two two NUM bracis-28410 223 8 images image NOUN bracis-28410 223 9 often often ADV bracis-28410 223 10 assume assume VERB bracis-28410 223 11 that that SCONJ bracis-28410 223 12 the the DET bracis-28410 223 13 images image NOUN bracis-28410 223 14 are be AUX bracis-28410 223 15 shown show VERB bracis-28410 223 16 side side NOUN bracis-28410 223 17 by by ADP bracis-28410 223 18 side side NOUN bracis-28410 223 19 or or CCONJ bracis-28410 223 20 alternated alternate VERB bracis-28410 223 21 with with ADP bracis-28410 223 22 an an DET bracis-28410 223 23 empty empty ADJ bracis-28410 223 24 image image NOUN bracis-28410 223 25 displayed display VERB bracis-28410 223 26 in in ADP bracis-28410 223 27 between between ADP bracis-28410 223 28 for for ADP bracis-28410 223 29 a a DET bracis-28410 223 30 short short ADJ bracis-28410 223 31 period period NOUN bracis-28410 223 32 before before SCONJ bracis-28410 223 33 the the DET bracis-28410 223 34 next next ADJ bracis-28410 223 35 image image NOUN bracis-28410 223 36 is be AUX bracis-28410 223 37 shown show VERB bracis-28410 223 38 . . PUNCT bracis-28410 224 1 in in ADP bracis-28410 224 2 contrast contrast NOUN bracis-28410 224 3 , , PUNCT bracis-28410 224 4 flipping flip VERB bracis-28410 224 5 ( ( PUNCT bracis-28410 224 6 or or CCONJ bracis-28410 224 7 alternating alternate VERB bracis-28410 224 8 ) ) PUNCT bracis-28410 224 9 between between ADP bracis-28410 224 10 two two NUM bracis-28410 224 11 similar similar ADJ bracis-28410 224 12 images image NOUN bracis-28410 224 13 reveals reveal VERB bracis-28410 224 14 their their PRON bracis-28410 224 15 differences difference NOUN bracis-28410 224 16 to to ADP bracis-28410 224 17 an an DET bracis-28410 224 18 observer observer NOUN bracis-28410 224 19 much much ADV bracis-28410 224 20 more more ADV bracis-28410 224 21 effectively effectively ADV bracis-28410 224 22 than than ADP bracis-28410 224 23 showing show VERB bracis-28410 224 24 the the DET bracis-28410 224 25 images image NOUN bracis-28410 224 26 next next ADV bracis-28410 224 27 to to ADP bracis-28410 224 28 each each DET bracis-28410 224 29 other other ADJ bracis-28410 224 30 . . PUNCT bracis-28410 225 1 aiming aim VERB bracis-28410 225 2 to to ADP bracis-28410 225 3 better well ADJ bracis-28410 225 4 approximate approximate ADJ bracis-28410 225 5 human human ADJ bracis-28410 225 6 evaluators evaluator NOUN bracis-28410 225 7 ’ ’ PART bracis-28410 225 8 methods method NOUN bracis-28410 225 9 , , PUNCT bracis-28410 225 10 andersson andersson PROPN bracis-28410 225 11 et et PROPN bracis-28410 225 12 . . PUNCT bracis-28410 226 1 al al PROPN bracis-28410 226 2 . . PROPN bracis-28410 226 3     SPACE bracis-28410 227 1 [ [ X bracis-28410 227 2 3 3 X bracis-28410 227 3 ] ] PUNCT bracis-28410 227 4 developed develop VERB bracis-28410 227 5 a a DET bracis-28410 227 6 full full ADJ bracis-28410 227 7 - - PUNCT bracis-28410 227 8 reference reference NOUN bracis-28410 227 9 image image NOUN bracis-28410 227 10 difference difference NOUN bracis-28410 227 11 algorithm algorithm NOUN bracis-28410 227 12 , , PUNCT bracis-28410 227 13 namely namely ADV bracis-28410 227 14     SPACE bracis-28410 227 15 , , PUNCT bracis-28410 227 16 which which PRON bracis-28410 227 17 carefully carefully ADV bracis-28410 227 18 evaluates evaluate VERB bracis-28410 227 19 differences difference NOUN bracis-28410 227 20 inspired inspire VERB bracis-28410 227 21 by by ADP bracis-28410 227 22 models model NOUN bracis-28410 227 23 of of ADP bracis-28410 227 24 the the DET bracis-28410 227 25 human human ADJ bracis-28410 227 26 vision vision NOUN bracis-28410 227 27 . . PUNCT bracis-28410 228 1 table table NOUN bracis-28410 228 2 1 1 NUM bracis-28410 228 3 . . PUNCT bracis-28410 228 4 model model NOUN bracis-28410 228 5 metrics metric NOUN bracis-28410 228 6 for for ADP bracis-28410 228 7 each each DET bracis-28410 228 8 evaluated evaluate VERB bracis-28410 228 9 dataset dataset NOUN bracis-28410 228 10 . . PUNCT bracis-28410 229 1 for for ADP bracis-28410 229 2 the the DET bracis-28410 229 3 psnr psnr NOUN bracis-28410 229 4 , , PUNCT bracis-28410 229 5 ssim ssim NOUN bracis-28410 229 6 , , PUNCT bracis-28410 229 7 and and CCONJ bracis-28410 229 8 ms ms ADJ bracis-28410 229 9 - - PUNCT bracis-28410 229 10 ssim ssim NOUN bracis-28410 229 11 metrics metric NOUN bracis-28410 229 12 the the ADV bracis-28410 229 13 higher high ADJ bracis-28410 229 14 the the PRON bracis-28410 229 15 better well ADJ bracis-28410 229 16 , , PUNCT bracis-28410 229 17 while while SCONJ bracis-28410 229 18 for for ADP bracis-28410 229 19 the the DET bracis-28410 229 20 metric metric NOUN bracis-28410 229 21 the the PRON bracis-28410 229 22 lower low ADJ bracis-28410 229 23 the the DET bracis-28410 229 24 better.full better.full NUM bracis-28410 229 25 size size NOUN bracis-28410 229 26 table table NOUN bracis-28410 229 27     SPACE bracis-28410 229 28 is be AUX bracis-28410 229 29 designed design VERB bracis-28410 229 30 to to PART bracis-28410 229 31 have have VERB bracis-28410 229 32 both both CCONJ bracis-28410 229 33 low low ADJ bracis-28410 229 34 complexity complexity NOUN bracis-28410 229 35 and and CCONJ bracis-28410 229 36 ease ease NOUN bracis-28410 229 37 of of ADP bracis-28410 229 38 use use NOUN bracis-28410 229 39 . . PUNCT bracis-28410 230 1 it it PRON bracis-28410 230 2 not not PART bracis-28410 230 3 only only ADV bracis-28410 230 4 evaluates evaluate VERB bracis-28410 230 5 differences difference NOUN bracis-28410 230 6 in in ADP bracis-28410 230 7 colors color NOUN bracis-28410 230 8 and and CCONJ bracis-28410 230 9 edges edge NOUN bracis-28410 230 10 , , PUNCT bracis-28410 230 11 but but CCONJ bracis-28410 230 12 also also ADV bracis-28410 230 13 pays pay VERB bracis-28410 230 14 attention attention NOUN bracis-28410 230 15 to to ADP bracis-28410 230 16 discrepancies discrepancy NOUN bracis-28410 230 17 in in ADP bracis-28410 230 18 isolated isolated ADJ bracis-28410 230 19 pixels pixel NOUN bracis-28410 230 20 with with ADP bracis-28410 230 21 colors color NOUN bracis-28410 230 22 that that PRON bracis-28410 230 23 greatly greatly ADV bracis-28410 230 24 differ differ VERB bracis-28410 230 25 from from ADP bracis-28410 230 26 their their PRON bracis-28410 230 27 surroundings surrounding NOUN bracis-28410 230 28 . . PUNCT bracis-28410 231 1 outputs output VERB bracis-28410 231 2 a a DET bracis-28410 231 3 new new ADJ bracis-28410 231 4 image image NOUN bracis-28410 231 5 indicating indicate VERB bracis-28410 231 6 the the DET bracis-28410 231 7 magnitude magnitude NOUN bracis-28410 231 8 of of ADP bracis-28410 231 9 the the DET bracis-28410 231 10 perceived perceive VERB bracis-28410 231 11 difference difference NOUN bracis-28410 231 12 between between ADP bracis-28410 231 13 two two NUM bracis-28410 231 14 images image NOUN bracis-28410 231 15 at at ADP bracis-28410 231 16 every every DET bracis-28410 231 17 pixel pixel NOUN bracis-28410 231 18 . . PUNCT bracis-28410 232 1 the the DET bracis-28410 232 2 algorithm algorithm NOUN bracis-28410 232 3 can can AUX bracis-28410 232 4 also also ADV bracis-28410 232 5 pool pool VERB bracis-28410 232 6 the the DET bracis-28410 232 7 per per ADP bracis-28410 232 8 - - PUNCT bracis-28410 232 9 pixel pixel NOUN bracis-28410 232 10 differences difference NOUN bracis-28410 232 11 down down ADP bracis-28410 232 12 to to ADP bracis-28410 232 13 a a DET bracis-28410 232 14 weighted weight VERB bracis-28410 232 15 histogram histogram NOUN bracis-28410 232 16 , , PUNCT bracis-28410 232 17 or or CCONJ bracis-28410 232 18 generate generate VERB bracis-28410 232 19 a a DET bracis-28410 232 20 single single ADJ bracis-28410 232 21 value value NOUN bracis-28410 232 22 , , PUNCT bracis-28410 232 23 which which PRON bracis-28410 232 24 is be AUX bracis-28410 232 25 the the DET bracis-28410 232 26 approach approach NOUN bracis-28410 232 27 we we PRON bracis-28410 232 28 will will AUX bracis-28410 232 29 use use VERB bracis-28410 232 30 during during ADP bracis-28410 232 31 our our PRON bracis-28410 232 32 analysis analysis NOUN bracis-28410 232 33 . . PUNCT bracis-28410 233 1 this this DET bracis-28410 233 2 value value NOUN bracis-28410 233 3 is be AUX bracis-28410 233 4 zero zero NUM bracis-28410 233 5 when when SCONJ bracis-28410 233 6 both both DET bracis-28410 233 7 images image NOUN bracis-28410 233 8 are be AUX bracis-28410 233 9 the the DET bracis-28410 233 10 same same ADJ bracis-28410 233 11 , , PUNCT bracis-28410 233 12 and and CCONJ bracis-28410 233 13 it it PRON bracis-28410 233 14 increases increase VERB bracis-28410 233 15 as as ADP bracis-28410 233 16 the the DET bracis-28410 233 17 more more ADV bracis-28410 233 18 noticeable noticeable ADJ bracis-28410 233 19 are be AUX bracis-28410 233 20 the the DET bracis-28410 233 21 differences difference NOUN bracis-28410 233 22 . . PUNCT bracis-28410 234 1 computational computational ADJ bracis-28410 234 2 resources resource NOUN bracis-28410 234 3 . . PUNCT bracis-28410 235 1 the the DET bracis-28410 235 2 super super ADJ bracis-28410 235 3 - - ADJ bracis-28410 235 4 resolution resolution ADJ bracis-28410 235 5 capsules capsule NOUN bracis-28410 235 6 ( ( PUNCT bracis-28410 235 7 srcaps srcap NOUN bracis-28410 235 8 ) ) PUNCT bracis-28410 235 9 model model NOUN bracis-28410 235 10 was be AUX bracis-28410 235 11 implemented implement VERB bracis-28410 235 12 using use VERB bracis-28410 235 13 the the DET bracis-28410 235 14 pytorch pytorch NOUN bracis-28410 235 15 open open ADJ bracis-28410 235 16 - - PUNCT bracis-28410 235 17 source source NOUN bracis-28410 235 18 platform platform NOUN bracis-28410 235 19 on on ADP bracis-28410 235 20 the the DET bracis-28410 235 21 pytorch pytorch NOUN bracis-28410 235 22 lightning lightning NOUN bracis-28410 235 23 wrapper wrapper NOUN bracis-28410 235 24 . . PUNCT bracis-28410 236 1 the the DET bracis-28410 236 2 implementation implementation NOUN bracis-28410 236 3 made make VERB bracis-28410 236 4 available available ADJ bracis-28410 236 5 by by ADP bracis-28410 236 6 uchidafootnote uchidafootnote NOUN bracis-28410 236 7 1 1 NUM bracis-28410 236 8 was be AUX bracis-28410 236 9 used use VERB bracis-28410 236 10 as as ADP bracis-28410 236 11 a a DET bracis-28410 236 12 foundation foundation NOUN bracis-28410 236 13 , , PUNCT bracis-28410 236 14 and and CCONJ bracis-28410 236 15 also also ADV bracis-28410 236 16 metrics metric NOUN bracis-28410 236 17 from from ADP bracis-28410 236 18 the the DET bracis-28410 236 19 pytorch pytorch NOUN bracis-28410 236 20 image image NOUN bracis-28410 236 21 quality quality NOUN bracis-28410 236 22 collection collection NOUN bracis-28410 236 23     SPACE bracis-28410 237 1 [ [ X bracis-28410 237 2 23 23 NUM bracis-28410 237 3 ] ] PUNCT bracis-28410 237 4 and and CCONJ bracis-28410 237 5 the the DET bracis-28410 237 6 official official ADJ bracis-28410 237 7 implementation implementation NOUN bracis-28410 237 8 of of ADP bracis-28410 237 9 the the DET bracis-28410 237 10 metric metric ADJ bracis-28410 237 11     SPACE bracis-28410 238 1 [ [ X bracis-28410 238 2 3 3 NUM bracis-28410 238 3 ] ] PUNCT bracis-28410 238 4 . . PUNCT bracis-28410 239 1 for for ADP bracis-28410 239 2 optimizers optimizer NOUN bracis-28410 239 3 implementations implementation NOUN bracis-28410 239 4 , , PUNCT bracis-28410 239 5 we we PRON bracis-28410 239 6 used use VERB bracis-28410 239 7 the the DET bracis-28410 239 8 torch torch NOUN bracis-28410 239 9 - - PUNCT bracis-28410 239 10 optimizer optimizer NOUN bracis-28410 239 11 packagefootnote packagefootnote NOUN bracis-28410 239 12 2 2 NUM bracis-28410 239 13 and and CCONJ bracis-28410 239 14 for for ADP bracis-28410 239 15 comparative comparative ADJ bracis-28410 239 16 visualization visualization NOUN bracis-28410 239 17 of of ADP bracis-28410 239 18 the the DET bracis-28410 239 19 metrics metric NOUN bracis-28410 239 20 and and CCONJ bracis-28410 239 21 generation generation NOUN bracis-28410 239 22 of of ADP bracis-28410 239 23 graphs graph NOUN bracis-28410 239 24 we we PRON bracis-28410 239 25 employed employ VERB bracis-28410 239 26 tensorboard tensorboard NOUN bracis-28410 239 27     SPACE bracis-28410 240 1 [ [ X bracis-28410 240 2 1 1 X bracis-28410 240 3 ] ] PUNCT bracis-28410 240 4 and and CCONJ bracis-28410 240 5 comet.mlfootnote comet.mlfootnote DET bracis-28410 240 6 3 3 NUM bracis-28410 240 7 tools tool NOUN bracis-28410 240 8 . . PUNCT bracis-28410 241 1 parameter parameter NOUN bracis-28410 241 2 search search NOUN bracis-28410 241 3 . . PUNCT bracis-28410 242 1 to to PART bracis-28410 242 2 find find VERB bracis-28410 242 3 the the DET bracis-28410 242 4 best good ADJ bracis-28410 242 5 set set NOUN bracis-28410 242 6 of of ADP bracis-28410 242 7 parameters parameter NOUN bracis-28410 242 8 for for ADP bracis-28410 242 9 our our PRON bracis-28410 242 10 srcaps srcap NOUN bracis-28410 242 11 model model NOUN bracis-28410 242 12 , , PUNCT bracis-28410 242 13 we we PRON bracis-28410 242 14 used use VERB bracis-28410 242 15 the the DET bracis-28410 242 16 ray ray NOUN bracis-28410 242 17     SPACE bracis-28410 243 1 [ [ X bracis-28410 243 2 32 32 NUM bracis-28410 243 3 ] ] X bracis-28410 243 4 open open ADJ bracis-28410 243 5 - - PUNCT bracis-28410 243 6 source source NOUN bracis-28410 243 7 framework framework NOUN bracis-28410 243 8 ( ( PUNCT bracis-28410 243 9 tune tune NOUN bracis-28410 243 10 ) ) PUNCT bracis-28410 243 11 , , PUNCT bracis-28410 243 12 which which PRON bracis-28410 243 13 is be AUX bracis-28410 243 14 a a DET bracis-28410 243 15 scalable scalable ADJ bracis-28410 243 16 hyperparameter hyperparameter NOUN bracis-28410 243 17 tuning tune VERB bracis-28410 243 18 library library NOUN bracis-28410 243 19 built build VERB bracis-28410 243 20 on on ADP bracis-28410 243 21 top top NOUN bracis-28410 243 22 of of ADP bracis-28410 243 23 ray ray NOUN bracis-28410 243 24 corefootnote corefootnote VERB bracis-28410 243 25 4 4 NUM bracis-28410 243 26 . . PUNCT bracis-28410 244 1 we we PRON bracis-28410 244 2 employed employ VERB bracis-28410 244 3 the the DET bracis-28410 244 4 async async NOUN bracis-28410 244 5 successive successive ADJ bracis-28410 244 6 halving halve VERB bracis-28410 244 7 ( ( PUNCT bracis-28410 244 8 asha asha PROPN bracis-28410 244 9 ) ) PUNCT bracis-28410 244 10 scheduler scheduler NOUN bracis-28410 244 11 during during ADP bracis-28410 244 12 the the DET bracis-28410 244 13 search search NOUN bracis-28410 244 14 , , PUNCT bracis-28410 244 15 as as SCONJ bracis-28410 244 16 it it PRON bracis-28410 244 17 decides decide VERB bracis-28410 244 18 at at ADP bracis-28410 244 19 each each DET bracis-28410 244 20 iteration iteration NOUN bracis-28410 244 21 which which PRON bracis-28410 244 22 trials trial NOUN bracis-28410 244 23 are be AUX bracis-28410 244 24 likely likely ADJ bracis-28410 244 25 to to PART bracis-28410 244 26 perform perform VERB bracis-28410 244 27 badly badly ADV bracis-28410 244 28 , , PUNCT bracis-28410 244 29 and and CCONJ bracis-28410 244 30 stops stop VERB bracis-28410 244 31 these these DET bracis-28410 244 32 trials trial NOUN bracis-28410 244 33 , , PUNCT bracis-28410 244 34 avoiding avoid VERB bracis-28410 244 35 wasting waste VERB bracis-28410 244 36 resources resource NOUN bracis-28410 244 37 on on ADP bracis-28410 244 38 poor poor ADJ bracis-28410 244 39 hyperparameter hyperparameter NOUN bracis-28410 244 40 configurations configuration NOUN bracis-28410 244 41 . . PUNCT bracis-28410 245 1 we we PRON bracis-28410 245 2 train train VERB bracis-28410 245 3 the the DET bracis-28410 245 4 models model NOUN bracis-28410 245 5 for for ADP bracis-28410 245 6 a a DET bracis-28410 245 7 maximum maximum NOUN bracis-28410 245 8 of of ADP bracis-28410 245 9 100 100 NUM bracis-28410 245 10 epochs epoch NOUN bracis-28410 245 11 in in ADP bracis-28410 245 12 the the DET bracis-28410 245 13 div2k div2k PROPN bracis-28410 245 14 training training NOUN bracis-28410 245 15 dataset dataset NOUN bracis-28410 245 16 and and CCONJ bracis-28410 245 17 evaluate evaluate VERB bracis-28410 245 18 them they PRON bracis-28410 245 19 using use VERB bracis-28410 245 20 the the DET bracis-28410 245 21 ms ms ADJ bracis-28410 245 22 - - PUNCT bracis-28410 245 23 ssim ssim NOUN bracis-28410 245 24 performance performance NOUN bracis-28410 245 25 on on ADP bracis-28410 245 26 the the DET bracis-28410 245 27 first first ADJ bracis-28410 245 28 10 10 NUM bracis-28410 245 29 images image NOUN bracis-28410 245 30 from from ADP bracis-28410 245 31 the the DET bracis-28410 245 32 div2k div2k PROPN bracis-28410 245 33 validation validation NOUN bracis-28410 245 34 dataset dataset NOUN bracis-28410 245 35 . . PUNCT bracis-28410 246 1 we we PRON bracis-28410 246 2 select select VERB bracis-28410 246 3 these these DET bracis-28410 246 4 values value NOUN bracis-28410 246 5 to to PART bracis-28410 246 6 allow allow VERB bracis-28410 246 7 fast fast ADJ bracis-28410 246 8 experimentation experimentation NOUN bracis-28410 246 9 of of ADP bracis-28410 246 10 different different ADJ bracis-28410 246 11 sets set NOUN bracis-28410 246 12 of of ADP bracis-28410 246 13 parameters parameter NOUN bracis-28410 246 14 , , PUNCT bracis-28410 246 15 as as SCONJ bracis-28410 246 16 it it PRON bracis-28410 246 17 was be AUX bracis-28410 246 18 observed observe VERB bracis-28410 246 19 during during ADP bracis-28410 246 20 numerous numerous ADJ bracis-28410 246 21 training training NOUN bracis-28410 246 22 processes process NOUN bracis-28410 246 23 that that PRON bracis-28410 246 24 the the DET bracis-28410 246 25 model model NOUN bracis-28410 246 26 tends tend VERB bracis-28410 246 27 to to PART bracis-28410 246 28 start start VERB bracis-28410 246 29 stabilizing stabilize VERB bracis-28410 246 30 performance performance NOUN bracis-28410 246 31 at at ADP bracis-28410 246 32 around around ADV bracis-28410 246 33 100 100 NUM bracis-28410 246 34 epochs epoch NOUN bracis-28410 246 35 . . PUNCT bracis-28410 247 1 experimental experimental ADJ bracis-28410 247 2 setup setup NOUN bracis-28410 247 3 . . PUNCT bracis-28410 248 1 during during ADP bracis-28410 248 2 the the DET bracis-28410 248 3 training training NOUN bracis-28410 248 4 process process NOUN bracis-28410 248 5 of of ADP bracis-28410 248 6 the the DET bracis-28410 248 7 different different ADJ bracis-28410 248 8 models model NOUN bracis-28410 248 9 used use VERB bracis-28410 248 10 for for ADP bracis-28410 248 11 comparison comparison NOUN bracis-28410 248 12 , , PUNCT bracis-28410 248 13 the the DET bracis-28410 248 14 entries entry NOUN bracis-28410 248 15 have have VERB bracis-28410 248 16 \(n \(n NOUN bracis-28410 248 17 = = SYM bracis-28410 248 18 16\ 16\ PROPN bracis-28410 248 19 ) ) PUNCT bracis-28410 248 20 ( ( PUNCT bracis-28410 248 21 batch batch NOUN bracis-28410 248 22 size size NOUN bracis-28410 248 23 ) ) PUNCT bracis-28410 248 24 pairs pair NOUN bracis-28410 248 25 of of ADP bracis-28410 248 26 image image NOUN bracis-28410 248 27 clippings clipping NOUN bracis-28410 248 28 from from ADP bracis-28410 248 29 the the DET bracis-28410 248 30 dataset dataset NOUN bracis-28410 248 31 . . PUNCT bracis-28410 249 1 for for ADP bracis-28410 249 2 all all DET bracis-28410 249 3 models model NOUN bracis-28410 249 4 , , PUNCT bracis-28410 249 5 during during ADP bracis-28410 249 6 validation validation NOUN bracis-28410 249 7 , , PUNCT bracis-28410 249 8 the the DET bracis-28410 249 9 lr lr NOUN bracis-28410 249 10 and and CCONJ bracis-28410 249 11 hr hr NOUN bracis-28410 249 12 images image NOUN bracis-28410 249 13 are be AUX bracis-28410 249 14 used use VERB bracis-28410 249 15 entirely entirely ADV bracis-28410 249 16 one one NUM bracis-28410 249 17 by by ADP bracis-28410 249 18 one one NUM bracis-28410 249 19 ( ( PUNCT bracis-28410 249 20 \(n \(n NOUN bracis-28410 249 21 = = SYM bracis-28410 249 22 1\ 1\ NUM bracis-28410 249 23 ) ) PUNCT bracis-28410 249 24 ) ) PUNCT bracis-28410 249 25 . . PUNCT bracis-28410 250 1 all all DET bracis-28410 250 2 models model NOUN bracis-28410 250 3 discussed discuss VERB bracis-28410 250 4 here here ADV bracis-28410 250 5 were be AUX bracis-28410 250 6 evaluated evaluate VERB bracis-28410 250 7 for for ADP bracis-28410 250 8 \(4\times \(4\time NOUN bracis-28410 250 9 \ \ NOUN bracis-28410 250 10 ) ) PUNCT bracis-28410 250 11 super super NOUN bracis-28410 250 12 - - NOUN bracis-28410 250 13 resolution resolution NOUN bracis-28410 250 14 for for ADP bracis-28410 250 15 2000 2000 NUM bracis-28410 250 16 epochs epoch NOUN bracis-28410 250 17 , , PUNCT bracis-28410 250 18 where where SCONJ bracis-28410 250 19 each each DET bracis-28410 250 20 epoch epoch NOUN bracis-28410 250 21 involves involve VERB bracis-28410 250 22 only only ADV bracis-28410 250 23 one one NUM bracis-28410 250 24 iteration iteration NOUN bracis-28410 250 25 through through ADP bracis-28410 250 26 the the DET bracis-28410 250 27 training training NOUN bracis-28410 250 28 dataset dataset NOUN bracis-28410 250 29 , , PUNCT bracis-28410 250 30 and and CCONJ bracis-28410 250 31 is be AUX bracis-28410 250 32 trained train VERB bracis-28410 250 33 with with ADP bracis-28410 250 34 two two NUM bracis-28410 250 35 different different ADJ bracis-28410 250 36 loss loss NOUN bracis-28410 250 37 functions function NOUN bracis-28410 250 38 : : PUNCT bracis-28410 250 39 l1 l1 PROPN bracis-28410 250 40 and and CCONJ bracis-28410 250 41 adaptive adaptive ADJ bracis-28410 250 42 . . PUNCT bracis-28410 251 1 other other ADJ bracis-28410 251 2 functions function NOUN bracis-28410 251 3 have have AUX bracis-28410 251 4 been be AUX bracis-28410 251 5 evaluated evaluate VERB bracis-28410 251 6 , , PUNCT bracis-28410 251 7 and and CCONJ bracis-28410 251 8 their their PRON bracis-28410 251 9 results result NOUN bracis-28410 251 10 will will AUX bracis-28410 251 11 be be AUX bracis-28410 251 12 briefly briefly ADV bracis-28410 251 13 discussed discuss VERB bracis-28410 251 14 throughout throughout ADP bracis-28410 251 15 this this DET bracis-28410 251 16 section section NOUN bracis-28410 251 17 . . PUNCT bracis-28410 252 1 the the DET bracis-28410 252 2 final final ADJ bracis-28410 252 3 srcaps srcap NOUN bracis-28410 252 4 model model NOUN bracis-28410 252 5 has have VERB bracis-28410 252 6 convolutional convolutional ADJ bracis-28410 252 7 kernels kernel NOUN bracis-28410 252 8 with with ADP bracis-28410 252 9 \(k \(k NOUN bracis-28410 252 10 = = SYM bracis-28410 252 11 3\ 3\ NUM bracis-28410 252 12 ) ) PUNCT bracis-28410 252 13 in in ADP bracis-28410 252 14 its its PRON bracis-28410 252 15 first first ADJ bracis-28410 252 16 layer layer NOUN bracis-28410 252 17 , , PUNCT bracis-28410 252 18 followed follow VERB bracis-28410 252 19 by by ADP bracis-28410 252 20 7 7 NUM bracis-28410 252 21 rdcbs rdcbs ADJ bracis-28410 252 22 ( ( PUNCT bracis-28410 252 23 \(b \(b X bracis-28410 252 24 = = SYM bracis-28410 252 25 7\ 7\ NOUN bracis-28410 252 26 ) ) PUNCT bracis-28410 252 27 ) ) PUNCT bracis-28410 252 28 . . PUNCT bracis-28410 253 1 the the DET bracis-28410 253 2 value value NOUN bracis-28410 253 3 used use VERB bracis-28410 253 4 for for ADP bracis-28410 253 5 the the DET bracis-28410 253 6 hyperparameters hyperparameter NOUN bracis-28410 253 7 l l NOUN bracis-28410 253 8 and and CCONJ bracis-28410 253 9 c c PROPN bracis-28410 253 10 are be AUX bracis-28410 253 11 the the DET bracis-28410 253 12 same same ADJ bracis-28410 253 13 for for ADP bracis-28410 253 14 all all DET bracis-28410 253 15 blocks block NOUN bracis-28410 253 16 : : PUNCT bracis-28410 253 17 3 3 NUM bracis-28410 253 18 and and CCONJ bracis-28410 253 19 4 4 NUM bracis-28410 253 20 , , PUNCT bracis-28410 253 21 respectively respectively ADV bracis-28410 253 22 . . PUNCT bracis-28410 254 1 in in ADP bracis-28410 254 2 the the DET bracis-28410 254 3 last last ADJ bracis-28410 254 4 convolutional convolutional ADJ bracis-28410 254 5 layer layer NOUN bracis-28410 254 6 , , PUNCT bracis-28410 254 7 we we PRON bracis-28410 254 8 chose choose VERB bracis-28410 254 9 \(k \(k NOUN bracis-28410 254 10 = = SYM bracis-28410 254 11 3\ 3\ NUM bracis-28410 254 12 ) ) PUNCT bracis-28410 254 13 , , PUNCT bracis-28410 254 14 as as ADV bracis-28410 254 15 well well ADV bracis-28410 254 16 as as ADP bracis-28410 254 17 for for ADP bracis-28410 254 18 the the DET bracis-28410 254 19 convolutional convolutional ADJ bracis-28410 254 20 layers layer NOUN bracis-28410 254 21 internal internal ADJ bracis-28410 254 22 to to ADP bracis-28410 254 23 the the DET bracis-28410 254 24 upnet upnet NOUN bracis-28410 254 25 . . PUNCT bracis-28410 255 1 throughout throughout ADP bracis-28410 255 2 the the DET bracis-28410 255 3 neural neural ADJ bracis-28410 255 4 network network NOUN bracis-28410 255 5 , , PUNCT bracis-28410 255 6 we we PRON bracis-28410 255 7 used use VERB bracis-28410 255 8 the the DET bracis-28410 255 9 values value NOUN bracis-28410 255 10 of of ADP bracis-28410 255 11 \(act \(act PROPN bracis-28410 255 12 = = PUNCT bracis-28410 255 13 relu\ relu\ PROPN bracis-28410 255 14 ) ) PUNCT bracis-28410 255 15 , , PUNCT bracis-28410 255 16 \(k \(k NOUN bracis-28410 255 17 = = SYM bracis-28410 255 18 3\ 3\ NUM bracis-28410 255 19 ) ) PUNCT bracis-28410 255 20 , , PUNCT bracis-28410 255 21 \(f \(f X bracis-28410 255 22 = = SYM bracis-28410 255 23 128\ 128\ NUM bracis-28410 255 24 ) ) PUNCT bracis-28410 255 25 , , PUNCT bracis-28410 255 26 \(st \(st NOUN bracis-28410 255 27 = = PROPN bracis-28410 255 28 1\ 1\ NUM bracis-28410 255 29 ) ) PUNCT bracis-28410 255 30 and and CCONJ bracis-28410 255 31 \(p \(p X bracis-28410 255 32 = = PUNCT bracis-28410 256 1 \text \text NOUN bracis-28410 256 2 { { PUNCT bracis-28410 256 3 ' ' PUNCT bracis-28410 256 4 same'}\ same'}\ PROPN bracis-28410 256 5 ) ) PUNCT bracis-28410 256 6 . . PUNCT bracis-28410 257 1 setting set VERB bracis-28410 257 2 the the DET bracis-28410 257 3 padding padding NOUN bracis-28410 257 4 to to ADP bracis-28410 257 5 the the DET bracis-28410 257 6 ’ ' PUNCT bracis-28410 257 7 same same ADJ bracis-28410 257 8 ’ ' PUNCT bracis-28410 257 9 mode mode NOUN bracis-28410 257 10 means mean VERB bracis-28410 257 11 using use VERB bracis-28410 257 12 its its PRON bracis-28410 257 13 value value NOUN bracis-28410 257 14 so so SCONJ bracis-28410 257 15 that that SCONJ bracis-28410 257 16 the the DET bracis-28410 257 17 input input NOUN bracis-28410 257 18 ( ( PUNCT bracis-28410 257 19 h h NOUN bracis-28410 257 20 and and CCONJ bracis-28410 257 21 w w NOUN bracis-28410 257 22 ) ) PUNCT bracis-28410 257 23 dimensions dimension NOUN bracis-28410 257 24 are be AUX bracis-28410 257 25 preserved preserve VERB bracis-28410 257 26 in in ADP bracis-28410 257 27 the the DET bracis-28410 257 28 output output NOUN bracis-28410 257 29 , , PUNCT bracis-28410 257 30 that that ADV bracis-28410 257 31 is is ADV bracis-28410 257 32 , , PUNCT bracis-28410 257 33 \(p \(p NOUN bracis-28410 257 34 = = SYM bracis-28410 257 35 \frac{k-1}{2}\ \frac{k-1}{2}\ NOUN bracis-28410 257 36 ) ) PUNCT bracis-28410 257 37 . . PUNCT bracis-28410 258 1 although although SCONJ bracis-28410 258 2 having have VERB bracis-28410 258 3 a a DET bracis-28410 258 4 smaller small ADJ bracis-28410 258 5 number number NOUN bracis-28410 258 6 of of ADP bracis-28410 258 7 layers layer NOUN bracis-28410 258 8 and and CCONJ bracis-28410 258 9 only only ADV bracis-28410 258 10 seven seven NUM bracis-28410 258 11 residual residual ADJ bracis-28410 258 12 blocks block NOUN bracis-28410 258 13 in in ADP bracis-28410 258 14 its its PRON bracis-28410 258 15 composition composition NOUN bracis-28410 258 16 , , PUNCT bracis-28410 258 17 the the DET bracis-28410 258 18 srcaps srcap NOUN bracis-28410 258 19 network network NOUN bracis-28410 258 20 has have VERB bracis-28410 258 21 a a DET bracis-28410 258 22 considerable considerable ADJ bracis-28410 258 23 number number NOUN bracis-28410 258 24 of of ADP bracis-28410 258 25 parameters parameter NOUN bracis-28410 258 26 15 15 NUM bracis-28410 258 27 m. m. NOUN bracis-28410 258 28 such such DET bracis-28410 258 29 a a DET bracis-28410 258 30 value value NOUN bracis-28410 258 31 represents represent VERB bracis-28410 258 32 13.5 13.5 NUM bracis-28410 258 33 m m NOUN bracis-28410 258 34 , , PUNCT bracis-28410 258 35 2.4 2.4 NUM bracis-28410 258 36 and and CCONJ bracis-28410 258 37 10.2 10.2 NUM bracis-28410 258 38 more more ADJ bracis-28410 258 39 parameters parameter NOUN bracis-28410 258 40 than than ADP bracis-28410 258 41 edsr edsr PROPN bracis-28410 258 42 , , PUNCT bracis-28410 258 43 rcan rcan ADJ bracis-28410 258 44 and and CCONJ bracis-28410 258 45 wdsr wdsr VERB bracis-28410 258 46 , , PUNCT bracis-28410 258 47 respectively respectively ADV bracis-28410 258 48 . . PUNCT bracis-28410 259 1 this this PRON bracis-28410 259 2 is be AUX bracis-28410 259 3 due due ADJ bracis-28410 259 4 to to ADP bracis-28410 259 5 the the DET bracis-28410 259 6 vectorial vectorial ADJ bracis-28410 259 7 nature nature NOUN bracis-28410 259 8 of of ADP bracis-28410 259 9 the the DET bracis-28410 259 10 capsule capsule NOUN bracis-28410 259 11 , , PUNCT bracis-28410 259 12 which which PRON bracis-28410 259 13 adds add VERB bracis-28410 259 14 an an DET bracis-28410 259 15 extra extra ADJ bracis-28410 259 16 dimension dimension NOUN bracis-28410 259 17 to to ADP bracis-28410 259 18 its its PRON bracis-28410 259 19 composition composition NOUN bracis-28410 259 20 . . PUNCT bracis-28410 260 1 hr hr NOUN bracis-28410 260 2 image image NOUN bracis-28410 260 3 slices slice NOUN bracis-28410 260 4 ( ( PUNCT bracis-28410 260 5 \(patch\_size\ \(patch\_size\ NOUN bracis-28410 260 6 ) ) PUNCT bracis-28410 260 7 ) ) PUNCT bracis-28410 260 8 of of ADP bracis-28410 260 9 size size NOUN bracis-28410 260 10 128\(\times 128\(\times NUM bracis-28410 260 11 \)128 \)128 PROPN bracis-28410 260 12 and and CCONJ bracis-28410 260 13 its its PRON bracis-28410 260 14 corresponding correspond VERB bracis-28410 260 15 32\(\times 32\(\times NUM bracis-28410 260 16 \)32 \)32 NOUN bracis-28410 260 17 lr lr NOUN bracis-28410 260 18 slice slice NOUN bracis-28410 260 19 were be AUX bracis-28410 260 20 used use VERB bracis-28410 260 21 during during ADP bracis-28410 260 22 their their PRON bracis-28410 260 23 training training NOUN bracis-28410 260 24 . . PUNCT bracis-28410 261 1 the the DET bracis-28410 261 2 updating updating NOUN bracis-28410 261 3 of of ADP bracis-28410 261 4 the the DET bracis-28410 261 5 weights weight NOUN bracis-28410 261 6 of of ADP bracis-28410 261 7 the the DET bracis-28410 261 8 networks network NOUN bracis-28410 261 9 is be AUX bracis-28410 261 10 done do VERB bracis-28410 261 11 by by ADP bracis-28410 261 12 the the DET bracis-28410 261 13 adam adam PROPN bracis-28410 261 14 optimizer optimizer NOUN bracis-28410 261 15     SPACE bracis-28410 262 1 [ [ X bracis-28410 262 2 27 27 NUM bracis-28410 262 3 ] ] PUNCT bracis-28410 262 4 , , PUNCT bracis-28410 262 5 with with ADP bracis-28410 262 6 \(\beta \(\beta PROPN bracis-28410 262 7 _ _ NOUN bracis-28410 262 8 { { PUNCT bracis-28410 262 9 1 1 NUM bracis-28410 262 10 } } PUNCT bracis-28410 262 11 = = SYM bracis-28410 262 12 0.9\ 0.9\ NOUN bracis-28410 262 13 ) ) PUNCT bracis-28410 262 14 , , PUNCT bracis-28410 262 15 \(\beta \(\beta X bracis-28410 262 16 _ _ NOUN bracis-28410 262 17 { { PUNCT bracis-28410 262 18 2 2 NUM bracis-28410 262 19 } } PUNCT bracis-28410 262 20 = = SYM bracis-28410 262 21 0.999\ 0.999\ NOUN bracis-28410 262 22 ) ) PUNCT bracis-28410 262 23 , , PUNCT bracis-28410 262 24 and and CCONJ bracis-28410 262 25 \(\epsilon \(\epsilon NOUN bracis-28410 262 26 = = NOUN bracis-28410 262 27 10^{-8}\ 10^{-8}\ NUM bracis-28410 262 28 ) ) PUNCT bracis-28410 262 29 , , PUNCT bracis-28410 262 30 being be AUX bracis-28410 262 31 the the DET bracis-28410 262 32 networks network NOUN bracis-28410 262 33 trained train VERB bracis-28410 262 34 with with ADP bracis-28410 262 35 an an DET bracis-28410 262 36 initial initial ADJ bracis-28410 262 37 learning learning NOUN bracis-28410 262 38 rate rate NOUN bracis-28410 262 39 of of ADP bracis-28410 262 40 \(lr \(lr NOUN bracis-28410 262 41 = = SYM bracis-28410 262 42 10^{-4}\ 10^{-4}\ NUM bracis-28410 262 43 ) ) PUNCT bracis-28410 262 44 which which PRON bracis-28410 262 45 decays decay VERB bracis-28410 262 46 to to ADP bracis-28410 262 47 half half NOUN bracis-28410 262 48 of of ADP bracis-28410 262 49 the the DET bracis-28410 262 50 current current ADJ bracis-28410 262 51 value value NOUN bracis-28410 262 52 every every DET bracis-28410 262 53 500 500 NUM bracis-28410 262 54 epochs epoch NOUN bracis-28410 262 55 . . PUNCT bracis-28410 263 1 when when SCONJ bracis-28410 263 2 used use VERB bracis-28410 263 3 , , PUNCT bracis-28410 263 4 the the DET bracis-28410 263 5 adaptive adaptive ADJ bracis-28410 263 6 loss loss NOUN bracis-28410 263 7 function function NOUN bracis-28410 263 8 was be AUX bracis-28410 263 9 initialized initialize VERB bracis-28410 263 10 with with ADP bracis-28410 263 11 the the DET bracis-28410 263 12 default default NOUN bracis-28410 263 13 values value NOUN bracis-28410 263 14 from from ADP bracis-28410 263 15 the the DET bracis-28410 263 16 official official ADJ bracis-28410 263 17 implementation implementation NOUN bracis-28410 263 18     SPACE bracis-28410 264 1 [ [ X bracis-28410 264 2 5 5 NUM bracis-28410 264 3 ] ] PUNCT bracis-28410 264 4 . . PUNCT bracis-28410 265 1 employing employ VERB bracis-28410 265 2 these these DET bracis-28410 265 3 values value NOUN bracis-28410 265 4 is be AUX bracis-28410 265 5 equivalent equivalent ADJ bracis-28410 265 6 to to ADP bracis-28410 265 7 starting start VERB bracis-28410 265 8 the the DET bracis-28410 265 9 training training NOUN bracis-28410 265 10 with with ADP bracis-28410 265 11 the the DET bracis-28410 265 12 charbonnier charbonnier NOUN bracis-28410 265 13 / / SYM bracis-28410 265 14 pseudo pseudo NOUN bracis-28410 265 15 - - NOUN bracis-28410 265 16 huber huber NOUN bracis-28410 265 17 function function NOUN bracis-28410 265 18 and and CCONJ bracis-28410 265 19 letting let VERB bracis-28410 265 20 the the DET bracis-28410 265 21 network network NOUN bracis-28410 265 22 learn learn VERB bracis-28410 265 23 from from ADP bracis-28410 265 24 it it PRON bracis-28410 265 25 what what DET bracis-28410 265 26 values value NOUN bracis-28410 265 27 for for ADP bracis-28410 265 28 its its PRON bracis-28410 265 29 parameters parameter NOUN bracis-28410 265 30 and and CCONJ bracis-28410 265 31 , , PUNCT bracis-28410 265 32 consequently consequently ADV bracis-28410 265 33 , , PUNCT bracis-28410 265 34 which which DET bracis-28410 265 35 function function NOUN bracis-28410 265 36 of of ADP bracis-28410 265 37 the the DET bracis-28410 265 38 subset subset NOUN bracis-28410 265 39 of of ADP bracis-28410 265 40 the the DET bracis-28410 265 41 general general ADJ bracis-28410 265 42 function function NOUN bracis-28410 265 43 is be AUX bracis-28410 265 44 more more ADV bracis-28410 265 45 appropriate appropriate ADJ bracis-28410 265 46 . . PUNCT bracis-28410 266 1 baselines baseline NOUN bracis-28410 266 2 and and CCONJ bracis-28410 266 3 results result NOUN bracis-28410 266 4 . . PUNCT bracis-28410 267 1 the the DET bracis-28410 267 2 edsr edsr PROPN bracis-28410 267 3 model model NOUN bracis-28410 267 4 used use VERB bracis-28410 267 5 is be AUX bracis-28410 267 6 the the DET bracis-28410 267 7 base base NOUN bracis-28410 267 8 model model NOUN bracis-28410 267 9 defined define VERB bracis-28410 267 10 by by ADP bracis-28410 267 11 lim lim PROPN bracis-28410 267 12 et et PROPN bracis-28410 267 13 al al PROPN bracis-28410 267 14 . . PUNCT bracis-28410 267 15     SPACE bracis-28410 268 1 [ [ X bracis-28410 268 2 33 33 NUM bracis-28410 268 3 ] ] PUNCT bracis-28410 268 4 . . PUNCT bracis-28410 269 1 we we PRON bracis-28410 269 2 chose choose VERB bracis-28410 269 3 the the DET bracis-28410 269 4 simplest simple ADJ bracis-28410 269 5 version version NOUN bracis-28410 269 6 of of ADP bracis-28410 269 7 the the DET bracis-28410 269 8 model model NOUN bracis-28410 269 9 because because SCONJ bracis-28410 269 10 it it PRON bracis-28410 269 11 has have VERB bracis-28410 269 12 a a DET bracis-28410 269 13 smaller small ADJ bracis-28410 269 14 number number NOUN bracis-28410 269 15 of of ADP bracis-28410 269 16 parameters parameter NOUN bracis-28410 269 17 than than ADP bracis-28410 269 18 the the DET bracis-28410 269 19 srcaps srcap NOUN bracis-28410 269 20 model model NOUN bracis-28410 269 21 . . PUNCT bracis-28410 270 1 it it PRON bracis-28410 270 2 is be AUX bracis-28410 270 3 composed compose VERB bracis-28410 270 4 of of ADP bracis-28410 270 5 16 16 NUM bracis-28410 270 6 residual residual ADJ bracis-28410 270 7 blocks block NOUN bracis-28410 270 8 without without ADP bracis-28410 270 9 residual residual ADJ bracis-28410 270 10 scale scale NOUN bracis-28410 270 11 application application NOUN bracis-28410 270 12 since since SCONJ bracis-28410 270 13 only only ADV bracis-28410 270 14 64 64 NUM bracis-28410 270 15 feature feature NOUN bracis-28410 270 16 maps map NOUN bracis-28410 270 17 ( ( PUNCT bracis-28410 270 18 filters filter NOUN bracis-28410 270 19 ) ) PUNCT bracis-28410 270 20 are be AUX bracis-28410 270 21 used use VERB bracis-28410 270 22 per per ADP bracis-28410 270 23 convolutional convolutional ADJ bracis-28410 270 24 layer layer NOUN bracis-28410 270 25 . . PUNCT bracis-28410 271 1 all all DET bracis-28410 271 2 convolutions convolution NOUN bracis-28410 271 3 , , PUNCT bracis-28410 271 4 including include VERB bracis-28410 271 5 the the DET bracis-28410 271 6 ones one NOUN bracis-28410 271 7 internal internal ADJ bracis-28410 271 8 to to ADP bracis-28410 271 9 the the DET bracis-28410 271 10 upscale upscale ADJ bracis-28410 271 11 network network NOUN bracis-28410 271 12 , , PUNCT bracis-28410 271 13 have have VERB bracis-28410 271 14 a a DET bracis-28410 271 15 kernel kernel NOUN bracis-28410 271 16 size size NOUN bracis-28410 271 17 of of ADP bracis-28410 271 18 3\(\times 3\(\times PROPN bracis-28410 271 19 \)3 \)3 PROPN bracis-28410 271 20 . . PUNCT bracis-28410 272 1 during during ADP bracis-28410 272 2 its its PRON bracis-28410 272 3 execution execution NOUN bracis-28410 272 4 , , PUNCT bracis-28410 272 5 the the DET bracis-28410 272 6 input input NOUN bracis-28410 272 7 images image NOUN bracis-28410 272 8 are be AUX bracis-28410 272 9 subtracted subtract VERB bracis-28410 272 10 from from ADP bracis-28410 272 11 the the DET bracis-28410 272 12 mean mean ADJ bracis-28410 272 13 rgb rgb PROPN bracis-28410 272 14 values value NOUN bracis-28410 272 15 of of ADP bracis-28410 272 16 the the DET bracis-28410 272 17 training training NOUN bracis-28410 272 18 images image NOUN bracis-28410 272 19 of of ADP bracis-28410 272 20 the the DET bracis-28410 272 21 div2k div2k PROPN bracis-28410 272 22 dataset dataset NOUN bracis-28410 272 23 , , PUNCT bracis-28410 272 24 which which PRON bracis-28410 272 25 are be AUX bracis-28410 272 26 0.4488 0.4488 NUM bracis-28410 272 27 , , PUNCT bracis-28410 272 28 0.4371 0.4371 NUM bracis-28410 272 29 and and CCONJ bracis-28410 272 30 0.4040 0.4040 NUM bracis-28410 272 31 . . PUNCT bracis-28410 273 1 these these DET bracis-28410 273 2 values value NOUN bracis-28410 273 3 range range VERB bracis-28410 273 4 from from ADP bracis-28410 273 5 0 0 NUM bracis-28410 273 6 to to ADP bracis-28410 273 7 1 1 NUM bracis-28410 273 8 and and CCONJ bracis-28410 273 9 are be AUX bracis-28410 273 10 multiplied multiply VERB bracis-28410 273 11 by by ADP bracis-28410 273 12 the the DET bracis-28410 273 13 maximum maximum PROPN bracis-28410 273 14 pixel pixel PROPN bracis-28410 273 15 value value NOUN bracis-28410 273 16 , , PUNCT bracis-28410 273 17 255 255 NUM bracis-28410 273 18 . . PUNCT bracis-28410 274 1 the the DET bracis-28410 274 2 rdn rdn NOUN bracis-28410 274 3     SPACE bracis-28410 274 4 [ [ PUNCT bracis-28410 274 5 64 64 NUM bracis-28410 274 6 ] ] PUNCT bracis-28410 274 7 model model NOUN bracis-28410 274 8 , , PUNCT bracis-28410 274 9 as as ADV bracis-28410 274 10 well well ADV bracis-28410 274 11 as as ADP bracis-28410 274 12 the the DET bracis-28410 274 13 edsr edsr NOUN bracis-28410 274 14 , , PUNCT bracis-28410 274 15 generates generate VERB bracis-28410 274 16 64 64 NUM bracis-28410 274 17 filters filter NOUN bracis-28410 274 18 as as ADP bracis-28410 274 19 the the DET bracis-28410 274 20 output output NOUN bracis-28410 274 21 of of ADP bracis-28410 274 22 its its PRON bracis-28410 274 23 convolutional convolutional ADJ bracis-28410 274 24 layers layer NOUN bracis-28410 274 25 , , PUNCT bracis-28410 274 26 and and CCONJ bracis-28410 274 27 uses use VERB bracis-28410 274 28 \(k \(k NOUN bracis-28410 274 29 = = SYM bracis-28410 274 30 3\ 3\ NUM bracis-28410 274 31 ) ) PUNCT bracis-28410 274 32 for for ADP bracis-28410 274 33 all all DET bracis-28410 274 34 convolutional convolutional ADJ bracis-28410 274 35 kernels kernel NOUN bracis-28410 274 36 , , PUNCT bracis-28410 274 37 except except SCONJ bracis-28410 274 38 for for ADP bracis-28410 274 39 those those PRON bracis-28410 274 40 used use VERB bracis-28410 274 41 in in ADP bracis-28410 274 42 the the DET bracis-28410 274 43 fusion fusion NOUN bracis-28410 274 44 layers layer NOUN bracis-28410 274 45 of of ADP bracis-28410 274 46 lff lff PROPN bracis-28410 274 47 and and CCONJ bracis-28410 274 48 gff gff PROPN bracis-28410 274 49 , , PUNCT bracis-28410 274 50 which which PRON bracis-28410 274 51 have have VERB bracis-28410 274 52 \(k \(k NOUN bracis-28410 274 53 = = SYM bracis-28410 274 54 1\ 1\ NUM bracis-28410 274 55 ) ) PUNCT bracis-28410 274 56 . . PUNCT bracis-28410 275 1 in in ADP bracis-28410 275 2 this this DET bracis-28410 275 3 model model NOUN bracis-28410 275 4 , , PUNCT bracis-28410 275 5 16 16 NUM bracis-28410 275 6 rdb rdb NOUN bracis-28410 275 7 blocks block NOUN bracis-28410 275 8 were be AUX bracis-28410 275 9 used use VERB bracis-28410 275 10 , , PUNCT bracis-28410 275 11 with with ADP bracis-28410 275 12 8 8 NUM bracis-28410 275 13 convolutional convolutional ADJ bracis-28410 275 14 layers layer NOUN bracis-28410 275 15 each each PRON bracis-28410 275 16 . . PUNCT bracis-28410 276 1 for for ADP bracis-28410 276 2 the the DET bracis-28410 276 3 wdsr wdsr PROPN bracis-28410 276 4 model model PROPN bracis-28410 276 5 , , PUNCT bracis-28410 276 6 we we PRON bracis-28410 276 7 used use VERB bracis-28410 276 8 the the DET bracis-28410 276 9 original original ADJ bracis-28410 276 10 implementation implementation NOUN bracis-28410 276 11 by by ADP bracis-28410 276 12 yu yu PROPN bracis-28410 276 13 et et PROPN bracis-28410 276 14 al al PROPN bracis-28410 276 15 . . PUNCT bracis-28410 276 16     SPACE bracis-28410 277 1 [ [ X bracis-28410 277 2 61 61 NUM bracis-28410 277 3 ] ] PUNCT bracis-28410 277 4 . . PUNCT bracis-28410 278 1 the the DET bracis-28410 278 2 chosen choose VERB bracis-28410 278 3 version version NOUN bracis-28410 278 4 was be AUX bracis-28410 278 5 wdsr wdsr ADJ bracis-28410 278 6 - - PUNCT bracis-28410 278 7 b b NOUN bracis-28410 278 8 , , PUNCT bracis-28410 278 9 with with ADP bracis-28410 278 10 16 16 NUM bracis-28410 278 11 large large ADJ bracis-28410 278 12 residual residual ADJ bracis-28410 278 13 blocks block NOUN bracis-28410 278 14 that that PRON bracis-28410 278 15 generate generate VERB bracis-28410 278 16 128 128 NUM bracis-28410 278 17 filters filter NOUN bracis-28410 278 18 , , PUNCT bracis-28410 278 19 but but CCONJ bracis-28410 278 20 which which PRON bracis-28410 278 21 internally internally ADV bracis-28410 278 22 generate generate VERB bracis-28410 278 23 convolutional convolutional ADJ bracis-28410 278 24 layers layer NOUN bracis-28410 278 25 with with ADP bracis-28410 278 26 \(6 \(6 NUM bracis-28410 278 27 \times \time NOUN bracis-28410 278 28 \ \ NOUN bracis-28410 278 29 ) ) PUNCT bracis-28410 278 30 more more ADJ bracis-28410 278 31 filters filter NOUN bracis-28410 278 32 . . PUNCT bracis-28410 279 1 this this DET bracis-28410 279 2 model model NOUN bracis-28410 279 3 , , PUNCT bracis-28410 279 4 such such ADJ bracis-28410 279 5 as as ADP bracis-28410 279 6 edsr edsr NOUN bracis-28410 279 7 , , PUNCT bracis-28410 279 8 also also ADV bracis-28410 279 9 subtracts subtract VERB bracis-28410 279 10 the the DET bracis-28410 279 11 mean mean PROPN bracis-28410 279 12 rgb rgb PROPN bracis-28410 279 13 values value NOUN bracis-28410 279 14 from from ADP bracis-28410 279 15 the the DET bracis-28410 279 16 div2k div2k PROPN bracis-28410 279 17 images image NOUN bracis-28410 279 18 . . PUNCT bracis-28410 280 1 the the DET bracis-28410 280 2 rcan rcan ADJ bracis-28410 280 3 model model NOUN bracis-28410 280 4 we we PRON bracis-28410 280 5 used use VERB bracis-28410 280 6 is be AUX bracis-28410 280 7 also also ADV bracis-28410 280 8 the the DET bracis-28410 280 9 original original ADJ bracis-28410 280 10 implementation implementation NOUN bracis-28410 280 11     SPACE bracis-28410 281 1 [ [ X bracis-28410 281 2 63 63 NUM bracis-28410 281 3 ] ] PUNCT bracis-28410 281 4 and and CCONJ bracis-28410 281 5 is be AUX bracis-28410 281 6 available available ADJ bracis-28410 281 7 in in ADP bracis-28410 281 8 the the DET bracis-28410 281 9 same same ADJ bracis-28410 281 10 repository repository NOUN bracis-28410 281 11 used use VERB bracis-28410 281 12 as as ADP bracis-28410 281 13 a a DET bracis-28410 281 14 baseline baseline NOUN bracis-28410 281 15 . . PUNCT bracis-28410 282 1 it it PRON bracis-28410 282 2 is be AUX bracis-28410 282 3 composed compose VERB bracis-28410 282 4 of of ADP bracis-28410 282 5 10 10 NUM bracis-28410 282 6 residual residual ADJ bracis-28410 282 7 groups group NOUN bracis-28410 282 8 ( ( PUNCT bracis-28410 282 9 rg rg NOUN bracis-28410 282 10 ) ) PUNCT bracis-28410 282 11 that that PRON bracis-28410 282 12 form form VERB bracis-28410 282 13 the the DET bracis-28410 282 14 residual residual ADJ bracis-28410 282 15 in in ADP bracis-28410 282 16 residual residual ADJ bracis-28410 282 17 ( ( PUNCT bracis-28410 282 18 rir rir PROPN bracis-28410 282 19 ) ) PUNCT bracis-28410 282 20 network network NOUN bracis-28410 282 21 structure structure NOUN bracis-28410 282 22 , , PUNCT bracis-28410 282 23 in in ADP bracis-28410 282 24 which which PRON bracis-28410 282 25 each each DET bracis-28410 282 26 rg rg PROPN bracis-28410 282 27 is be AUX bracis-28410 282 28 composed compose VERB bracis-28410 282 29 of of ADP bracis-28410 282 30 16 16 NUM bracis-28410 282 31 residual residual ADJ bracis-28410 282 32 channel channel NOUN bracis-28410 282 33 attention attention NOUN bracis-28410 282 34 blocks block NOUN bracis-28410 282 35 ( ( PUNCT bracis-28410 282 36 rcab rcab NOUN bracis-28410 282 37 ) ) PUNCT bracis-28410 282 38 . . PUNCT bracis-28410 283 1 it it PRON bracis-28410 283 2 has have VERB bracis-28410 283 3 \(k \(k NOUN bracis-28410 283 4 = = SYM bracis-28410 283 5 3\ 3\ NUM bracis-28410 283 6 ) ) PUNCT bracis-28410 283 7 kernel kernel NOUN bracis-28410 283 8 sizes size NOUN bracis-28410 283 9 which which PRON bracis-28410 283 10 generate generate VERB bracis-28410 283 11 \(c \(c NOUN bracis-28410 283 12 = = SYM bracis-28410 283 13 64\ 64\ NUM bracis-28410 283 14 ) ) PUNCT bracis-28410 283 15 filters filter NOUN bracis-28410 283 16 in in ADP bracis-28410 283 17 all all DET bracis-28410 283 18 convolutional convolutional ADJ bracis-28410 283 19 layers layer NOUN bracis-28410 283 20 , , PUNCT bracis-28410 283 21 except except SCONJ bracis-28410 283 22 for for ADP bracis-28410 283 23 those those PRON bracis-28410 283 24 in in ADP bracis-28410 283 25 the the DET bracis-28410 283 26 channel channel NOUN bracis-28410 283 27 reduction reduction NOUN bracis-28410 283 28 and and CCONJ bracis-28410 283 29 amplification amplification NOUN bracis-28410 283 30 mechanism mechanism NOUN bracis-28410 283 31 , , PUNCT bracis-28410 283 32 which which PRON bracis-28410 283 33 have have VERB bracis-28410 283 34 \(k \(k NOUN bracis-28410 283 35 = = SYM bracis-28410 283 36 1\ 1\ NUM bracis-28410 283 37 ) ) PUNCT bracis-28410 283 38 , , PUNCT bracis-28410 283 39 and and CCONJ bracis-28410 283 40 \(\frac{c}{r}=4\ \(\frac{c}{r}=4\ NOUN bracis-28410 283 41 ) ) PUNCT bracis-28410 283 42 and and CCONJ bracis-28410 283 43 \(c \(c NOUN bracis-28410 283 44 = = SYM bracis-28410 283 45 64\ 64\ NUM bracis-28410 283 46 ) ) PUNCT bracis-28410 283 47 respectively respectively ADV bracis-28410 283 48 , , PUNCT bracis-28410 283 49 with with ADP bracis-28410 283 50 reduction reduction NOUN bracis-28410 283 51 factor factor NOUN bracis-28410 283 52 \(r \(r NOUN bracis-28410 283 53 = = SYM bracis-28410 283 54 16\ 16\ NUM bracis-28410 283 55 ) ) PUNCT bracis-28410 283 56 . . PUNCT bracis-28410 284 1 table table NOUN bracis-28410 284 2     SPACE bracis-28410 284 3 1 1 NUM bracis-28410 284 4 summarizes summarize NOUN bracis-28410 284 5 the the DET bracis-28410 284 6 results result NOUN bracis-28410 284 7 obtained obtain VERB bracis-28410 284 8 for for ADP bracis-28410 284 9 all all DET bracis-28410 284 10 models model NOUN bracis-28410 284 11 and and CCONJ bracis-28410 284 12 datasets dataset NOUN bracis-28410 284 13 after after ADP bracis-28410 284 14 the the DET bracis-28410 284 15 learning learning NOUN bracis-28410 284 16 phase phase NOUN bracis-28410 284 17 . . PUNCT bracis-28410 285 1 from from ADP bracis-28410 285 2 this this DET bracis-28410 285 3 table table NOUN bracis-28410 285 4 , , PUNCT bracis-28410 285 5 we we PRON bracis-28410 285 6 highlight highlight VERB bracis-28410 285 7 the the DET bracis-28410 285 8 following follow VERB bracis-28410 285 9 points point NOUN bracis-28410 285 10 . . PUNCT bracis-28410 286 1 first first ADV bracis-28410 286 2 , , PUNCT bracis-28410 286 3 the the DET bracis-28410 286 4 srcaps srcap NOUN bracis-28410 286 5 model model NOUN bracis-28410 286 6 obtained obtain VERB bracis-28410 286 7 comparable comparable ADJ bracis-28410 286 8 results result NOUN bracis-28410 286 9 to to ADP bracis-28410 286 10 the the DET bracis-28410 286 11 edsr edsr PROPN bracis-28410 286 12 model model NOUN bracis-28410 286 13 , , PUNCT bracis-28410 286 14 sometimes sometimes ADV bracis-28410 286 15 surpassing surpass VERB bracis-28410 286 16 it it PRON bracis-28410 286 17 for for ADP bracis-28410 286 18 some some DET bracis-28410 286 19 metrics metric NOUN bracis-28410 286 20 , , PUNCT bracis-28410 286 21 particularly particularly ADV bracis-28410 286 22 in in ADP bracis-28410 286 23 the the DET bracis-28410 286 24 b100 b100 PROPN bracis-28410 286 25 dataset dataset NOUN bracis-28410 286 26 . . PUNCT bracis-28410 287 1 second second ADJ bracis-28410 287 2 , , PUNCT bracis-28410 287 3 as as SCONJ bracis-28410 287 4 shown show VERB bracis-28410 287 5 in in ADP bracis-28410 287 6 figs fig NOUN bracis-28410 287 7 . . PUNCT bracis-28410 287 8     SPACE bracis-28410 288 1 4 4 NUM bracis-28410 288 2 and and CCONJ bracis-28410 288 3     SPACE bracis-28410 288 4 5 5 NUM bracis-28410 288 5 , , PUNCT bracis-28410 288 6 the the DET bracis-28410 288 7 proposed propose VERB bracis-28410 288 8 model model NOUN bracis-28410 288 9 manages manage VERB bracis-28410 288 10 to to PART bracis-28410 288 11 recreate recreate VERB bracis-28410 288 12 the the DET bracis-28410 288 13 connection connection NOUN bracis-28410 288 14 between between ADP bracis-28410 288 15 the the DET bracis-28410 288 16 different different ADJ bracis-28410 288 17 characters character NOUN bracis-28410 288 18 more more ADV bracis-28410 288 19 precisely precisely ADV bracis-28410 288 20 , , PUNCT bracis-28410 288 21 while while SCONJ bracis-28410 288 22 models model NOUN bracis-28410 288 23 with with ADP bracis-28410 288 24 better well ADJ bracis-28410 288 25 metrics metric NOUN bracis-28410 288 26 such such ADJ bracis-28410 288 27 as as ADP bracis-28410 288 28 rcan rcan ADJ bracis-28410 288 29 and and CCONJ bracis-28410 288 30 rdn rdn NOUN bracis-28410 288 31 tend tend VERB bracis-28410 288 32 to to PART bracis-28410 288 33 thin thin VERB bracis-28410 288 34 the the DET bracis-28410 288 35 connection connection NOUN bracis-28410 288 36 , , PUNCT bracis-28410 288 37 as as SCONJ bracis-28410 288 38 they they PRON bracis-28410 288 39 do do VERB bracis-28410 288 40 for for ADP bracis-28410 288 41 the the DET bracis-28410 288 42 leftmost leftmost ADJ bracis-28410 288 43 part part NOUN bracis-28410 288 44 of of ADP bracis-28410 288 45 the the DET bracis-28410 288 46 symbol symbol NOUN bracis-28410 288 47 on on ADP bracis-28410 288 48 top top NOUN bracis-28410 288 49 . . PUNCT bracis-28410 289 1 finally finally ADV bracis-28410 289 2 , , PUNCT bracis-28410 289 3 despite despite SCONJ bracis-28410 289 4 being be AUX bracis-28410 289 5 able able ADJ bracis-28410 289 6 to to PART bracis-28410 289 7 reconstruct reconstruct VERB bracis-28410 289 8 with with ADP bracis-28410 289 9 quality quality NOUN bracis-28410 289 10 rounded rounded ADJ bracis-28410 289 11 edges edge NOUN bracis-28410 289 12 , , PUNCT bracis-28410 289 13 a a DET bracis-28410 289 14 deficiency deficiency NOUN bracis-28410 289 15 of of ADP bracis-28410 289 16 the the DET bracis-28410 289 17 srcaps srcap NOUN bracis-28410 289 18 model model NOUN bracis-28410 289 19 is be AUX bracis-28410 289 20 in in ADP bracis-28410 289 21 the the DET bracis-28410 289 22 reconstruction reconstruction NOUN bracis-28410 289 23 of of ADP bracis-28410 289 24 linear linear ADJ bracis-28410 289 25 edges edge NOUN bracis-28410 289 26 , , PUNCT bracis-28410 289 27 usually usually ADV bracis-28410 289 28 diagonal diagonal ADJ bracis-28410 289 29 . . PUNCT bracis-28410 290 1 it it PRON bracis-28410 290 2 is be AUX bracis-28410 290 3 remarkable remarkable ADJ bracis-28410 290 4 the the DET bracis-28410 290 5 results result NOUN bracis-28410 290 6 obtained obtain VERB bracis-28410 290 7 with with ADP bracis-28410 290 8 the the DET bracis-28410 290 9 rcan rcan ADJ bracis-28410 290 10 model model NOUN bracis-28410 290 11 , , PUNCT bracis-28410 290 12 reaching reach VERB bracis-28410 290 13 the the DET bracis-28410 290 14 highest high ADJ bracis-28410 290 15 value value NOUN bracis-28410 290 16 in in ADP bracis-28410 290 17 all all DET bracis-28410 290 18 metrics metric NOUN bracis-28410 290 19 . . PUNCT bracis-28410 291 1 we we PRON bracis-28410 291 2 highlight highlight VERB bracis-28410 291 3 that that SCONJ bracis-28410 291 4 our our PRON bracis-28410 291 5 goal goal NOUN bracis-28410 291 6 is be AUX bracis-28410 291 7 not not PART bracis-28410 291 8 to to PART bracis-28410 291 9 push push VERB bracis-28410 291 10 the the DET bracis-28410 291 11 state state NOUN bracis-28410 291 12 - - PUNCT bracis-28410 291 13 of of ADP bracis-28410 291 14 the the DET bracis-28410 291 15 art art NOUN bracis-28410 291 16 but but CCONJ bracis-28410 291 17 to to PART bracis-28410 291 18 bring bring VERB bracis-28410 291 19 insights insight NOUN bracis-28410 291 20 into into ADP bracis-28410 291 21 capsules capsule NOUN bracis-28410 291 22 applied apply VERB bracis-28410 291 23 to to ADP bracis-28410 291 24 super super ADJ bracis-28410 291 25 - - ADJ bracis-28410 291 26 resolution resolution ADJ bracis-28410 291 27 problems problem NOUN bracis-28410 291 28 . . PUNCT bracis-28410 292 1 we we PRON bracis-28410 292 2 believe believe VERB bracis-28410 292 3 that that SCONJ bracis-28410 292 4 future future ADJ bracis-28410 292 5 research research NOUN bracis-28410 292 6 on on ADP bracis-28410 292 7 capsule capsule NOUN bracis-28410 292 8 networks network NOUN bracis-28410 292 9 could could AUX bracis-28410 292 10 benefit benefit VERB bracis-28410 292 11 from from ADP bracis-28410 292 12 our our PRON bracis-28410 292 13 findings finding NOUN bracis-28410 292 14 . . PUNCT bracis-28410 293 1 fig fig NOUN bracis-28410 293 2 . . PUNCT bracis-28410 294 1 4 4 X bracis-28410 294 2 . . X bracis-28410 294 3 model model NOUN bracis-28410 294 4 results result NOUN bracis-28410 294 5 for for ADP bracis-28410 294 6 “ " PUNCT bracis-28410 294 7 0891 0891 NUM bracis-28410 294 8 ” " PUNCT bracis-28410 294 9 image image NOUN bracis-28410 294 10 from from ADP bracis-28410 294 11 div2k div2k PROPN bracis-28410 294 12 dataset dataset VERB bracis-28410 294 13 . . PUNCT bracis-28410 295 1 full full ADJ bracis-28410 295 2 size size NOUN bracis-28410 295 3 image image NOUN bracis-28410 295 4 fig fig NOUN bracis-28410 295 5 . . PUNCT bracis-28410 296 1 5 5 X bracis-28410 296 2 . . X bracis-28410 296 3 model model NOUN bracis-28410 296 4 results result NOUN bracis-28410 296 5 for for ADP bracis-28410 296 6 “ " PUNCT bracis-28410 296 7 0829 0829 NUM bracis-28410 296 8 ” " PUNCT bracis-28410 296 9 image image NOUN bracis-28410 296 10 from from ADP bracis-28410 296 11 div2k div2k PROPN bracis-28410 296 12 dataset dataset VERB bracis-28410 296 13 . . PUNCT bracis-28410 297 1 full full ADJ bracis-28410 297 2 size size NOUN bracis-28410 297 3 image image NOUN bracis-28410 297 4 5 5 NUM bracis-28410 297 5 conclusions conclusion NOUN bracis-28410 297 6 the the DET bracis-28410 297 7 purpose purpose NOUN bracis-28410 297 8 of of ADP bracis-28410 297 9 this this DET bracis-28410 297 10 work work NOUN bracis-28410 297 11 was be AUX bracis-28410 297 12 to to PART bracis-28410 297 13 evaluate evaluate VERB bracis-28410 297 14 the the DET bracis-28410 297 15 use use NOUN bracis-28410 297 16 of of ADP bracis-28410 297 17 the the DET bracis-28410 297 18 capsule capsule NOUN bracis-28410 297 19 concept concept NOUN bracis-28410 297 20 in in ADP bracis-28410 297 21 the the DET bracis-28410 297 22 solution solution NOUN bracis-28410 297 23 of of ADP bracis-28410 297 24 single single ADJ bracis-28410 297 25 image image NOUN bracis-28410 297 26 super super ADJ bracis-28410 297 27 - - ADJ bracis-28410 297 28 resolution resolution ADJ bracis-28410 297 29 problems problem NOUN bracis-28410 297 30 , , PUNCT bracis-28410 297 31 as as ADV bracis-28410 297 32 well well ADV bracis-28410 297 33 as as ADP bracis-28410 297 34 to to PART bracis-28410 297 35 verify verify VERB bracis-28410 297 36 new new ADJ bracis-28410 297 37 forms form NOUN bracis-28410 297 38 of of ADP bracis-28410 297 39 training training NOUN bracis-28410 297 40 and and CCONJ bracis-28410 297 41 validate validate VERB bracis-28410 297 42 the the DET bracis-28410 297 43 results result NOUN bracis-28410 297 44 of of ADP bracis-28410 297 45 neural neural ADJ bracis-28410 297 46 networks network NOUN bracis-28410 297 47 for for ADP bracis-28410 297 48 this this DET bracis-28410 297 49 purpose purpose NOUN bracis-28410 297 50 . . PUNCT bracis-28410 298 1 it it PRON bracis-28410 298 2 was be AUX bracis-28410 298 3 evidenced evidence VERB bracis-28410 298 4 that that SCONJ bracis-28410 298 5 , , PUNCT bracis-28410 298 6 despite despite SCONJ bracis-28410 298 7 the the DET bracis-28410 298 8 inferior inferior ADJ bracis-28410 298 9 result result NOUN bracis-28410 298 10 , , PUNCT bracis-28410 298 11 a a DET bracis-28410 298 12 trained train VERB bracis-28410 298 13 network network NOUN bracis-28410 298 14 with with ADP bracis-28410 298 15 a a DET bracis-28410 298 16 smaller small ADJ bracis-28410 298 17 number number NOUN bracis-28410 298 18 of of ADP bracis-28410 298 19 layers layer NOUN bracis-28410 298 20 obtained obtain VERB bracis-28410 298 21 a a DET bracis-28410 298 22 relevant relevant ADJ bracis-28410 298 23 result result NOUN bracis-28410 298 24 , , PUNCT bracis-28410 298 25 indicating indicate VERB bracis-28410 298 26 that that SCONJ bracis-28410 298 27 networks network NOUN bracis-28410 298 28 that that PRON bracis-28410 298 29 use use VERB bracis-28410 298 30 capsules capsule NOUN bracis-28410 298 31 can can AUX bracis-28410 298 32 have have VERB bracis-28410 298 33 applications application NOUN bracis-28410 298 34 in in ADP bracis-28410 298 35 super super NOUN bracis-28410 298 36 - - NOUN bracis-28410 298 37 resolution resolution NOUN bracis-28410 298 38 . . PUNCT bracis-28410 299 1 hypotheses hypothesis NOUN bracis-28410 299 2 have have AUX bracis-28410 299 3 been be AUX bracis-28410 299 4 raised raise VERB bracis-28410 299 5 that that SCONJ bracis-28410 299 6 the the DET bracis-28410 299 7 nonlinearity nonlinearity NOUN bracis-28410 299 8 function function NOUN bracis-28410 299 9 applied apply VERB bracis-28410 299 10 together together ADV bracis-28410 299 11 with with ADP bracis-28410 299 12 the the DET bracis-28410 299 13 capsules capsule NOUN bracis-28410 299 14 may may AUX bracis-28410 299 15 be be AUX bracis-28410 299 16 a a DET bracis-28410 299 17 limiting limit VERB bracis-28410 299 18 factor factor NOUN bracis-28410 299 19 , , PUNCT bracis-28410 299 20 given give VERB bracis-28410 299 21 the the DET bracis-28410 299 22 different different ADJ bracis-28410 299 23 nature nature NOUN bracis-28410 299 24 of of ADP bracis-28410 299 25 the the DET bracis-28410 299 26 problem problem NOUN bracis-28410 299 27 as as ADP bracis-28410 299 28 to to ADP bracis-28410 299 29 its its PRON bracis-28410 299 30 initial initial ADJ bracis-28410 299 31 usage usage NOUN bracis-28410 299 32 ( ( PUNCT bracis-28410 299 33 super super NOUN bracis-28410 299 34 - - NOUN bracis-28410 299 35 resolution resolution ADJ bracis-28410 299 36 \(\times \(\time NOUN bracis-28410 299 37 \ \ NOUN bracis-28410 299 38 ) ) PUNCT bracis-28410 299 39 classification classification NOUN bracis-28410 299 40 ) ) PUNCT bracis-28410 299 41 . . PUNCT bracis-28410 300 1 throughout throughout ADP bracis-28410 300 2 our our PRON bracis-28410 300 3 work work NOUN bracis-28410 300 4 , , PUNCT bracis-28410 300 5 we we PRON bracis-28410 300 6 investigate investigate VERB bracis-28410 300 7 the the DET bracis-28410 300 8 contribution contribution NOUN bracis-28410 300 9 of of ADP bracis-28410 300 10 many many ADJ bracis-28410 300 11 hyperparameters hyperparameter NOUN bracis-28410 300 12 such such ADJ bracis-28410 300 13 as as ADP bracis-28410 300 14 the the DET bracis-28410 300 15 activation activation NOUN bracis-28410 300 16 function function NOUN bracis-28410 300 17 , , PUNCT bracis-28410 300 18 learning learn VERB bracis-28410 300 19 rate rate NOUN bracis-28410 300 20 , , PUNCT bracis-28410 300 21 optimizer optimizer NOUN bracis-28410 300 22 and and CCONJ bracis-28410 300 23 loss loss NOUN bracis-28410 300 24 function function NOUN bracis-28410 300 25 . . PUNCT bracis-28410 301 1 regarding regard VERB bracis-28410 301 2 the the DET bracis-28410 301 3 latter latter ADJ bracis-28410 301 4 , , PUNCT bracis-28410 301 5 we we PRON bracis-28410 301 6 evaluate evaluate VERB bracis-28410 301 7 loss loss NOUN bracis-28410 301 8 functions function NOUN bracis-28410 301 9 that that PRON bracis-28410 301 10 take take VERB bracis-28410 301 11 the the DET bracis-28410 301 12 human human ADJ bracis-28410 301 13 visual visual ADJ bracis-28410 301 14 system system NOUN bracis-28410 301 15 into into ADP bracis-28410 301 16 account account NOUN bracis-28410 301 17 , , PUNCT bracis-28410 301 18 as as SCONJ bracis-28410 301 19 suggested suggest VERB bracis-28410 301 20 by by ADP bracis-28410 301 21 previous previous ADJ bracis-28410 301 22 works work NOUN bracis-28410 301 23     SPACE bracis-28410 302 1 [ [ X bracis-28410 302 2 55 55 NUM bracis-28410 302 3 , , PUNCT bracis-28410 302 4 57 57 NUM bracis-28410 302 5 ] ] PUNCT bracis-28410 302 6 . . PUNCT bracis-28410 303 1 additionally additionally ADV bracis-28410 303 2 , , PUNCT bracis-28410 303 3 we we PRON bracis-28410 303 4 study study VERB bracis-28410 303 5 different different ADJ bracis-28410 303 6 architectural architectural ADJ bracis-28410 303 7 designs design NOUN bracis-28410 303 8 to to PART bracis-28410 303 9 compose compose VERB bracis-28410 303 10 our our PRON bracis-28410 303 11 capsule capsule NOUN bracis-28410 303 12 network network NOUN bracis-28410 303 13 . . PUNCT bracis-28410 304 1 the the DET bracis-28410 304 2 fact fact NOUN bracis-28410 304 3 that that SCONJ bracis-28410 304 4 the the DET bracis-28410 304 5 adaptive adaptive ADJ bracis-28410 304 6 function function NOUN bracis-28410 304 7     SPACE bracis-28410 305 1 [ [ X bracis-28410 305 2 5 5 NUM bracis-28410 305 3 ] ] PUNCT bracis-28410 305 4 is be AUX bracis-28410 305 5 a a DET bracis-28410 305 6 superset superset NOUN bracis-28410 305 7 of of ADP bracis-28410 305 8 several several ADJ bracis-28410 305 9 others other NOUN bracis-28410 305 10 and and CCONJ bracis-28410 305 11 that that SCONJ bracis-28410 305 12 it it PRON bracis-28410 305 13 is be AUX bracis-28410 305 14 possible possible ADJ bracis-28410 305 15 to to PART bracis-28410 305 16 make make VERB bracis-28410 305 17 the the DET bracis-28410 305 18 network network NOUN bracis-28410 305 19 learn learn VERB bracis-28410 305 20 , , PUNCT bracis-28410 305 21 along along ADP bracis-28410 305 22 with with ADP bracis-28410 305 23 the the DET bracis-28410 305 24 other other ADJ bracis-28410 305 25 weights weight NOUN bracis-28410 305 26 , , PUNCT bracis-28410 305 27 the the DET bracis-28410 305 28 optimal optimal ADJ bracis-28410 305 29 values value NOUN bracis-28410 305 30 for for ADP bracis-28410 305 31 its its PRON bracis-28410 305 32 two two NUM bracis-28410 305 33 main main ADJ bracis-28410 305 34 parameters parameter NOUN bracis-28410 305 35 ( ( PUNCT bracis-28410 305 36 \(\alpha \(\alpha PROPN bracis-28410 305 37 \ \ NOUN bracis-28410 305 38 ) ) PUNCT bracis-28410 305 39 and and CCONJ bracis-28410 305 40 c c X bracis-28410 305 41 ) ) PUNCT bracis-28410 305 42 , , PUNCT bracis-28410 305 43 allow allow VERB bracis-28410 305 44 the the DET bracis-28410 305 45 network network NOUN bracis-28410 305 46 to to PART bracis-28410 305 47 experiment experiment VERB bracis-28410 305 48 which which DET bracis-28410 305 49 loss loss NOUN bracis-28410 305 50 function function NOUN bracis-28410 305 51 best well ADV bracis-28410 305 52 fits fit VERB bracis-28410 305 53 the the DET bracis-28410 305 54 problem problem NOUN bracis-28410 305 55 . . PUNCT bracis-28410 306 1 thus thus ADV bracis-28410 306 2 , , PUNCT bracis-28410 306 3 it it PRON bracis-28410 306 4 is be AUX bracis-28410 306 5 possible possible ADJ bracis-28410 306 6 to to PART bracis-28410 306 7 train train VERB bracis-28410 306 8 the the DET bracis-28410 306 9 network network NOUN bracis-28410 306 10 starting start VERB bracis-28410 306 11 from from ADP bracis-28410 306 12 a a DET bracis-28410 306 13 function function NOUN bracis-28410 306 14 similar similar ADJ bracis-28410 306 15 to to ADP bracis-28410 306 16 l1 l1 PROPN bracis-28410 306 17 , , PUNCT bracis-28410 306 18 while while SCONJ bracis-28410 306 19 modifying modify VERB bracis-28410 306 20 it it PRON bracis-28410 306 21 at at ADP bracis-28410 306 22 each each DET bracis-28410 306 23 iteration iteration NOUN bracis-28410 306 24 to to PART bracis-28410 306 25 extract extract VERB bracis-28410 306 26 as as ADV bracis-28410 306 27 much much ADJ bracis-28410 306 28 useful useful ADJ bracis-28410 306 29 information information NOUN bracis-28410 306 30 as as ADP bracis-28410 306 31 possible possible ADJ bracis-28410 306 32 from from ADP bracis-28410 306 33 the the DET bracis-28410 306 34 data datum NOUN bracis-28410 306 35 . . PUNCT bracis-28410 307 1 the the DET bracis-28410 307 2 current current ADJ bracis-28410 307 3 limitations limitation NOUN bracis-28410 307 4 of of ADP bracis-28410 307 5 the the DET bracis-28410 307 6 most most ADV bracis-28410 307 7 used use VERB bracis-28410 307 8 metrics metric NOUN bracis-28410 307 9 in in ADP bracis-28410 307 10 the the DET bracis-28410 307 11 literature literature NOUN bracis-28410 307 12     SPACE bracis-28410 307 13 [ [ X bracis-28410 307 14 31 31 NUM bracis-28410 307 15 ] ] PUNCT bracis-28410 307 16 were be AUX bracis-28410 307 17 also also ADV bracis-28410 307 18 emphasized emphasize VERB bracis-28410 307 19 in in ADP bracis-28410 307 20 this this DET bracis-28410 307 21 work work NOUN bracis-28410 307 22 , , PUNCT bracis-28410 307 23 showing show VERB bracis-28410 307 24 that that SCONJ bracis-28410 307 25 visual visual ADJ bracis-28410 307 26 evaluation evaluation NOUN bracis-28410 307 27 of of ADP bracis-28410 307 28 the the DET bracis-28410 307 29 results result NOUN bracis-28410 307 30 is be AUX bracis-28410 307 31 still still ADV bracis-28410 307 32 essential essential ADJ bracis-28410 307 33 . . PUNCT bracis-28410 308 1 existing exist VERB bracis-28410 308 2 metrics metric NOUN bracis-28410 308 3 have have AUX bracis-28410 308 4 been be AUX bracis-28410 308 5 suggested suggest VERB bracis-28410 308 6 such such ADJ bracis-28410 308 7 as as ADP bracis-28410 308 8 ms ms PROPN bracis-28410 308 9 - - PUNCT bracis-28410 308 10 ssim ssim NOUN bracis-28410 308 11     SPACE bracis-28410 309 1 [ [ X bracis-28410 309 2 55 55 NUM bracis-28410 309 3 ] ] PUNCT bracis-28410 309 4 and and CCONJ bracis-28410 309 5     SPACE bracis-28410 310 1 [ [ X bracis-28410 310 2 3 3 NUM bracis-28410 310 3 ] ] PUNCT bracis-28410 310 4 , , PUNCT bracis-28410 310 5 encouraging encourage VERB bracis-28410 310 6 discussion discussion NOUN bracis-28410 310 7 of of ADP bracis-28410 310 8 new new ADJ bracis-28410 310 9 metrics metric NOUN bracis-28410 310 10 . . PUNCT bracis-28410 311 1 several several ADJ bracis-28410 311 2 points point NOUN bracis-28410 311 3 of of ADP bracis-28410 311 4 possible possible ADJ bracis-28410 311 5 improvements improvement NOUN bracis-28410 311 6 that that PRON bracis-28410 311 7 could could AUX bracis-28410 311 8 not not PART bracis-28410 311 9 be be AUX bracis-28410 311 10 deeply deeply ADV bracis-28410 311 11 evaluated evaluate VERB bracis-28410 311 12 were be AUX bracis-28410 311 13 identified identify VERB bracis-28410 311 14 . . PUNCT bracis-28410 312 1 as as ADP bracis-28410 312 2 a a DET bracis-28410 312 3 future future ADJ bracis-28410 312 4 line line NOUN bracis-28410 312 5 of of ADP bracis-28410 312 6 research research NOUN bracis-28410 312 7 , , PUNCT bracis-28410 312 8 we we PRON bracis-28410 312 9 intend intend VERB bracis-28410 312 10 to to PART bracis-28410 312 11 replace replace VERB bracis-28410 312 12 the the DET bracis-28410 312 13 composition composition NOUN bracis-28410 312 14 of of ADP bracis-28410 312 15 the the DET bracis-28410 312 16 upnet upnet NOUN bracis-28410 312 17 network network NOUN bracis-28410 312 18 , , PUNCT bracis-28410 312 19 which which PRON bracis-28410 312 20 is be AUX bracis-28410 312 21 used use VERB bracis-28410 312 22 in in ADP bracis-28410 312 23 much much ADV bracis-28410 312 24 the the DET bracis-28410 312 25 same same ADJ bracis-28410 312 26 way way NOUN bracis-28410 312 27 by by ADP bracis-28410 312 28 several several ADJ bracis-28410 312 29 networks network NOUN bracis-28410 312 30 that that PRON bracis-28410 312 31 have have AUX bracis-28410 312 32 reached reach VERB bracis-28410 312 33 the the DET bracis-28410 312 34 state state NOUN bracis-28410 312 35 of of ADP bracis-28410 312 36 the the DET bracis-28410 312 37 art art NOUN bracis-28410 312 38 . . PUNCT bracis-28410 313 1 one one PRON bracis-28410 313 2 can can AUX bracis-28410 313 3 again again ADV bracis-28410 313 4 verify verify VERB bracis-28410 313 5 the the DET bracis-28410 313 6 usage usage NOUN bracis-28410 313 7 of of ADP bracis-28410 313 8 the the DET bracis-28410 313 9 concept concept NOUN bracis-28410 313 10 of of ADP bracis-28410 313 11 reverse reverse ADJ bracis-28410 313 12 convolutions convolution NOUN bracis-28410 313 13 , , PUNCT bracis-28410 313 14 or or CCONJ bracis-28410 313 15 deconvolutions deconvolution NOUN bracis-28410 313 16 , , PUNCT bracis-28410 313 17 as as SCONJ bracis-28410 313 18 used use VERB bracis-28410 313 19 by by ADP bracis-28410 313 20 dong dong PROPN bracis-28410 313 21 et et PROPN bracis-28410 313 22 al al PROPN bracis-28410 313 23 . . PUNCT bracis-28410 313 24     SPACE bracis-28410 314 1 [ [ X bracis-28410 314 2 12 12 NUM bracis-28410 314 3 ] ] PUNCT bracis-28410 314 4 , , PUNCT bracis-28410 314 5 and and CCONJ bracis-28410 314 6 also also ADV bracis-28410 314 7 of of ADP bracis-28410 314 8 deconvolutional deconvolutional ADJ bracis-28410 314 9 capsules capsule NOUN bracis-28410 314 10 created create VERB bracis-28410 314 11 by by ADP bracis-28410 314 12 lalonde lalonde NOUN bracis-28410 314 13 and and CCONJ bracis-28410 314 14 bagci bagci NOUN bracis-28410 314 15     SPACE bracis-28410 315 1 [ [ X bracis-28410 315 2 28 28 NUM bracis-28410 315 3 ] ] PUNCT bracis-28410 315 4 , , PUNCT bracis-28410 315 5 or or CCONJ bracis-28410 315 6 use use VERB bracis-28410 315 7 more more ADV bracis-28410 315 8 recent recent ADJ bracis-28410 315 9 methods method NOUN bracis-28410 315 10 from from ADP bracis-28410 315 11 the the DET bracis-28410 315 12 literature literature NOUN bracis-28410 315 13 . . PUNCT bracis-28410 316 1 kim kim PROPN bracis-28410 316 2 and and CCONJ bracis-28410 316 3 lee lee PROPN bracis-28410 316 4     SPACE bracis-28410 317 1 [ [ X bracis-28410 317 2 26 26 NUM bracis-28410 317 3 ] ] PUNCT bracis-28410 317 4 recently recently ADV bracis-28410 317 5 proposed propose VERB bracis-28410 317 6 the the DET bracis-28410 317 7 enhanced enhanced ADJ bracis-28410 317 8 upscaling upscale VERB bracis-28410 317 9 module module NOUN bracis-28410 317 10 ( ( PUNCT bracis-28410 317 11 eum eum PROPN bracis-28410 317 12 ) ) PUNCT bracis-28410 317 13 , , PUNCT bracis-28410 317 14 which which PRON bracis-28410 317 15 achieves achieve VERB bracis-28410 317 16 better well ADJ bracis-28410 317 17 results result NOUN bracis-28410 317 18 through through ADP bracis-28410 317 19 nonlinearities nonlinearitie NOUN bracis-28410 317 20 and and CCONJ bracis-28410 317 21 residual residual ADJ bracis-28410 317 22 connections connection NOUN bracis-28410 317 23 . . PUNCT bracis-28410 318 1 notes note VERB bracis-28410 318 2 1.github.com/s-aiueo32/sr-pytorch-lightning 1.github.com/s-aiueo32/sr-pytorch-lightning PRON bracis-28410 318 3 . . PUNCT bracis-28410 319 1 2.https://github.com/jettify/pytorch-optimizer 2.https://github.com/jettify/pytorch-optimizer NUM bracis-28410 319 2 . . PUNCT bracis-28410 320 1 3.https://www.comet.ml/. 3.https://www.comet.ml/. NUM bracis-28410 320 2 4.https://docs.ray.io/en/master/tune/index.html 4.https://docs.ray.io/en/master/tune/index.html NUM bracis-28410 320 3 . . PUNCT bracis-28410 321 1 references reference NOUN bracis-28410 321 2 abadi abadi PROPN bracis-28410 321 3 , , PUNCT bracis-28410 321 4 m. m. NOUN bracis-28410 321 5 , , PUNCT bracis-28410 321 6 et et PROPN bracis-28410 321 7 al al PROPN bracis-28410 321 8 . . PUNCT bracis-28410 321 9 : : PUNCT bracis-28410 322 1 tensorflow tensorflow NOUN bracis-28410 322 2 : : PUNCT bracis-28410 322 3 large large ADJ bracis-28410 322 4 - - PUNCT bracis-28410 322 5 scale scale NOUN bracis-28410 322 6 machine machine NOUN bracis-28410 322 7 learning learn VERB bracis-28410 322 8 on on ADP bracis-28410 322 9 heterogeneous heterogeneous ADJ bracis-28410 322 10 systems system NOUN bracis-28410 322 11 ( ( PUNCT bracis-28410 322 12 2015 2015 NUM bracis-28410 322 13 ) ) PUNCT bracis-28410 322 14 . . PUNCT bracis-28410 323 1 software software NOUN bracis-28410 323 2 available available ADJ bracis-28410 323 3 from from ADP bracis-28410 323 4 tensorflow.org tensorflow.org X bracis-28410 323 5 google google PROPN bracis-28410 323 6 scholar scholar NOUN bracis-28410 323 7     SPACE bracis-28410 323 8 agustsson agustsson NOUN bracis-28410 323 9 , , PUNCT bracis-28410 323 10 e. e. PROPN bracis-28410 323 11 , , PUNCT bracis-28410 323 12 timofte timofte PROPN bracis-28410 323 13 , , PUNCT bracis-28410 323 14 r. r. PROPN bracis-28410 323 15 : : PUNCT bracis-28410 323 16 ntire ntire NOUN bracis-28410 323 17 2017 2017 NUM bracis-28410 323 18 challenge challenge NOUN bracis-28410 323 19 on on ADP bracis-28410 323 20 single single ADJ bracis-28410 323 21 image image NOUN bracis-28410 323 22 super super NOUN bracis-28410 323 23 - - NOUN bracis-28410 323 24 resolution resolution NOUN bracis-28410 323 25 : : PUNCT bracis-28410 323 26 dataset dataset NOUN bracis-28410 323 27 and and CCONJ bracis-28410 323 28 study study NOUN bracis-28410 323 29 . . PUNCT bracis-28410 324 1 in in ADP bracis-28410 324 2 : : PUNCT bracis-28410 324 3 cvpr cvpr NOUN bracis-28410 324 4 workshops workshop NOUN bracis-28410 324 5 , , PUNCT bracis-28410 324 6 pp pp ADV bracis-28410 324 7 . . PUNCT bracis-28410 325 1 1–8 1–8 NUM bracis-28410 325 2 ( ( PUNCT bracis-28410 325 3 2017 2017 NUM bracis-28410 325 4 ) ) PUNCT bracis-28410 325 5 google google PROPN bracis-28410 325 6 scholar scholar NOUN bracis-28410 325 7     SPACE bracis-28410 325 8 andersson andersson PROPN bracis-28410 325 9 , , PUNCT bracis-28410 325 10 p. p. PROPN bracis-28410 325 11 , , PUNCT bracis-28410 325 12 nilsson nilsson PROPN bracis-28410 325 13 , , PUNCT bracis-28410 325 14 j. j. PROPN bracis-28410 325 15 , , PUNCT bracis-28410 325 16 akenine akenine NOUN bracis-28410 325 17 - - PUNCT bracis-28410 325 18 möller möller NOUN bracis-28410 325 19 , , PUNCT bracis-28410 325 20 t. t. PROPN bracis-28410 325 21 , , PUNCT bracis-28410 325 22 oskarsson oskarsson NOUN bracis-28410 325 23 , , PUNCT bracis-28410 325 24 m. m. NOUN bracis-28410 325 25 , , PUNCT bracis-28410 325 26 åström åström NOUN bracis-28410 325 27 , , PUNCT bracis-28410 325 28 k. k. PROPN bracis-28410 325 29 , , PUNCT bracis-28410 325 30 fairchild fairchild PROPN bracis-28410 325 31 , , PUNCT bracis-28410 325 32 m.d m.d PROPN bracis-28410 325 33 . . PUNCT bracis-28410 325 34 : : PUNCT bracis-28410 326 1 flip flip VERB bracis-28410 326 2 : : PUNCT bracis-28410 326 3 a a DET bracis-28410 326 4 difference difference NOUN bracis-28410 326 5 evaluator evaluator NOUN bracis-28410 326 6 for for ADP bracis-28410 326 7 alternating alternate VERB bracis-28410 326 8 images image NOUN bracis-28410 326 9 . . PUNCT bracis-28410 327 1 in in ADP bracis-28410 327 2 : : PUNCT bracis-28410 327 3 acm acm NOUN bracis-28410 327 4 on on ADP bracis-28410 327 5 computer computer NOUN bracis-28410 327 6 graphics graphic NOUN bracis-28410 327 7 and and CCONJ bracis-28410 327 8 interactive interactive ADJ bracis-28410 327 9 techniques technique NOUN bracis-28410 327 10 ( ( PUNCT bracis-28410 327 11 2020 2020 NUM bracis-28410 327 12 ) ) PUNCT bracis-28410 327 13 google google PROPN bracis-28410 327 14 scholar scholar NOUN bracis-28410 327 15     SPACE bracis-28410 327 16 anwar anwar PROPN bracis-28410 327 17 , , PUNCT bracis-28410 327 18 s. s. PROPN bracis-28410 327 19 , , PUNCT bracis-28410 327 20 khan khan PROPN bracis-28410 327 21 , , PUNCT bracis-28410 327 22 s.h s.h PROPN bracis-28410 327 23 . . PROPN bracis-28410 327 24 , , PUNCT bracis-28410 327 25 barnes barnes PROPN bracis-28410 327 26 , , PUNCT bracis-28410 327 27 n. n. NOUN bracis-28410 327 28 : : PUNCT bracis-28410 327 29 a a DET bracis-28410 327 30 deep deep ADJ bracis-28410 327 31 journey journey NOUN bracis-28410 327 32 into into ADP bracis-28410 327 33 super super NOUN bracis-28410 327 34 - - NOUN bracis-28410 327 35 resolution resolution NOUN bracis-28410 327 36 : : PUNCT bracis-28410 327 37 a a DET bracis-28410 327 38 survey survey NOUN bracis-28410 327 39 . . PUNCT bracis-28410 327 40 acm acm PROPN bracis-28410 327 41 comput comput NOUN bracis-28410 327 42 . . PUNCT bracis-28410 328 1 surv surv PROPN bracis-28410 328 2 . . PUNCT bracis-28410 329 1 53(3 53(3 NUM bracis-28410 329 2 ) ) PUNCT bracis-28410 329 3 , , PUNCT bracis-28410 329 4 60:1–60:34 60:1–60:34 NUM bracis-28410 329 5 ( ( PUNCT bracis-28410 329 6 2020 2020 NUM bracis-28410 329 7 ) ) PUNCT bracis-28410 329 8 . . PUNCT bracis-28410 330 1 https://doi.org/10.1145/3390462 https://doi.org/10.1145/3390462 PROPN bracis-28410 330 2 barron barron PROPN bracis-28410 330 3 , , PUNCT bracis-28410 330 4 j.t j.t PROPN bracis-28410 330 5 . . PROPN bracis-28410 330 6 : : PUNCT bracis-28410 330 7 a a DET bracis-28410 330 8 more more ADV bracis-28410 330 9 general general ADJ bracis-28410 330 10 robust robust ADJ bracis-28410 330 11 loss loss NOUN bracis-28410 330 12 function function NOUN bracis-28410 330 13 . . PUNCT bracis-28410 331 1 arxiv arxiv PROPN bracis-28410 331 2 preprint preprint VERB bracis-28410 331 3 arxiv:1701.03077 arxiv:1701.03077 PROPN bracis-28410 331 4 ( ( PUNCT bracis-28410 331 5 2017 2017 NUM bracis-28410 331 6 ) ) PUNCT bracis-28410 331 7 behjati behjati NOUN bracis-28410 331 8 , , PUNCT bracis-28410 331 9 p. p. NOUN bracis-28410 331 10 , , PUNCT bracis-28410 331 11 rodriguez rodriguez PROPN bracis-28410 331 12 , , PUNCT bracis-28410 331 13 p. p. PROPN bracis-28410 331 14 , , PUNCT bracis-28410 331 15 mehri mehri PROPN bracis-28410 331 16 , , PUNCT bracis-28410 331 17 a. a. NOUN bracis-28410 331 18 , , PUNCT bracis-28410 331 19 hupont hupont PROPN bracis-28410 331 20 , , PUNCT bracis-28410 331 21 i. i. PROPN bracis-28410 331 22 , , PUNCT bracis-28410 331 23 tena tena PROPN bracis-28410 331 24 , , PUNCT bracis-28410 331 25 c.f c.f PROPN bracis-28410 331 26 . . PROPN bracis-28410 331 27 , , PUNCT bracis-28410 331 28 gonzalez gonzalez PROPN bracis-28410 331 29 , , PUNCT bracis-28410 331 30 j. j. PROPN bracis-28410 331 31 : : PUNCT bracis-28410 331 32 overnet overnet NOUN bracis-28410 331 33 : : PUNCT bracis-28410 331 34 lightweight lightweight ADJ bracis-28410 331 35 multi multi ADJ bracis-28410 331 36 - - ADJ bracis-28410 331 37 scale scale ADJ bracis-28410 331 38 super super NOUN bracis-28410 331 39 - - NOUN bracis-28410 331 40 resolution resolution NOUN bracis-28410 331 41 with with ADP bracis-28410 331 42 overscaling overscale VERB bracis-28410 331 43 network network NOUN bracis-28410 331 44 . . PUNCT bracis-28410 332 1 in in ADP bracis-28410 332 2 : : PUNCT bracis-28410 332 3 wacv wacv ADJ bracis-28410 332 4 , , PUNCT bracis-28410 332 5 pp pp ADP bracis-28410 332 6 . . PUNCT bracis-28410 333 1 1–11 1–11 PROPN bracis-28410 333 2 ( ( PUNCT bracis-28410 333 3 2021 2021 NUM bracis-28410 333 4 ) ) PUNCT bracis-28410 333 5 google google NOUN bracis-28410 333 6 scholar scholar NOUN bracis-28410 333 7     SPACE bracis-28410 333 8 bevilacqua bevilacqua NOUN bracis-28410 333 9 , , PUNCT bracis-28410 333 10 m. m. NOUN bracis-28410 333 11 , , PUNCT bracis-28410 333 12 roumy roumy NOUN bracis-28410 333 13 , , PUNCT bracis-28410 333 14 a. a. NOUN bracis-28410 333 15 , , PUNCT bracis-28410 333 16 guillemot guillemot NOUN bracis-28410 333 17 , , PUNCT bracis-28410 333 18 c. c. PROPN bracis-28410 333 19 , , PUNCT bracis-28410 333 20 alberi alberi NOUN bracis-28410 333 21 - - PUNCT bracis-28410 333 22 morel morel PROPN bracis-28410 333 23 , , PUNCT bracis-28410 333 24 m.l m.l PROPN bracis-28410 333 25 . . PUNCT bracis-28410 333 26 : : PUNCT bracis-28410 333 27 low low ADJ bracis-28410 333 28 - - PUNCT bracis-28410 333 29 complexity complexity NOUN bracis-28410 333 30 single single ADJ bracis-28410 333 31 - - PUNCT bracis-28410 333 32 image image NOUN bracis-28410 333 33 super super NOUN bracis-28410 333 34 - - NOUN bracis-28410 333 35 resolution resolution NOUN bracis-28410 333 36 based base VERB bracis-28410 333 37 on on ADP bracis-28410 333 38 nonnegative nonnegative ADJ bracis-28410 333 39 neighbor neighbor NOUN bracis-28410 333 40 embedding embed VERB bracis-28410 333 41 . . PUNCT bracis-28410 334 1 in in ADP bracis-28410 334 2 : : PUNCT bracis-28410 334 3 bmvc bmvc NOUN bracis-28410 334 4 ( ( PUNCT bracis-28410 334 5 2012 2012 NUM bracis-28410 334 6 ) ) PUNCT bracis-28410 334 7 google google PROPN bracis-28410 334 8 scholar scholar NOUN bracis-28410 334 9     SPACE bracis-28410 334 10 cai cai PROPN bracis-28410 334 11 , , PUNCT bracis-28410 334 12 j. j. PROPN bracis-28410 334 13 , , PUNCT bracis-28410 334 14 gu gu PROPN bracis-28410 334 15 , , PUNCT bracis-28410 334 16 s. s. PROPN bracis-28410 334 17 , , PUNCT bracis-28410 334 18 timofte timofte PROPN bracis-28410 334 19 , , PUNCT bracis-28410 334 20 r. r. PROPN bracis-28410 334 21 , , PUNCT bracis-28410 334 22 zhang zhang PROPN bracis-28410 334 23 , , PUNCT bracis-28410 334 24 l. l. PROPN bracis-28410 334 25 : : PUNCT bracis-28410 334 26 ntire ntire NOUN bracis-28410 334 27 2019 2019 NUM bracis-28410 334 28 challenge challenge NOUN bracis-28410 334 29 on on ADP bracis-28410 334 30 real real ADJ bracis-28410 334 31 image image NOUN bracis-28410 334 32 super super NOUN bracis-28410 334 33 - - NOUN bracis-28410 334 34 resolution resolution NOUN bracis-28410 334 35 : : PUNCT bracis-28410 334 36 methods method NOUN bracis-28410 334 37 and and CCONJ bracis-28410 334 38 results result NOUN bracis-28410 334 39 . . PUNCT bracis-28410 335 1 in in ADP bracis-28410 335 2 : : PUNCT bracis-28410 335 3 cvpr cvpr NOUN bracis-28410 335 4 workshops workshop NOUN bracis-28410 335 5 , , PUNCT bracis-28410 335 6 pp pp ADV bracis-28410 335 7 . . PUNCT bracis-28410 336 1 1–8 1–8 NUM bracis-28410 336 2 ( ( PUNCT bracis-28410 336 3 2019 2019 NUM bracis-28410 336 4 ) ) PUNCT bracis-28410 336 5 google google NOUN bracis-28410 336 6 scholar scholar NOUN bracis-28410 336 7     SPACE bracis-28410 336 8 canny canny ADJ bracis-28410 336 9 , , PUNCT bracis-28410 336 10 j. j. PROPN bracis-28410 336 11 : : PUNCT bracis-28410 336 12 a a DET bracis-28410 336 13 computational computational ADJ bracis-28410 336 14 approach approach NOUN bracis-28410 336 15 to to ADP bracis-28410 336 16 edge edge NOUN bracis-28410 336 17 detection detection NOUN bracis-28410 336 18 . . PUNCT bracis-28410 337 1 trans tran NOUN bracis-28410 337 2 . . PUNCT bracis-28410 337 3 pattern pattern PROPN bracis-28410 337 4 anal anal PROPN bracis-28410 337 5 . . PUNCT bracis-28410 338 1 mach mach PROPN bracis-28410 338 2 . . PUNCT bracis-28410 339 1 intell intell PROPN bracis-28410 339 2 . . PUNCT bracis-28410 339 3 8(6 8(6 NUM bracis-28410 339 4 ) ) PUNCT bracis-28410 339 5 , , PUNCT bracis-28410 339 6 679–698 679–698 NUM bracis-28410 339 7 ( ( PUNCT bracis-28410 339 8 1986 1986 NUM bracis-28410 339 9 ) ) PUNCT bracis-28410 339 10 article article NOUN bracis-28410 339 11     SPACE bracis-28410 339 12 google google PROPN bracis-28410 339 13 scholar scholar NOUN bracis-28410 339 14     SPACE bracis-28410 339 15 dong dong PROPN bracis-28410 339 16 , , PUNCT bracis-28410 339 17 c. c. PROPN bracis-28410 339 18 , , PUNCT bracis-28410 339 19 loy loy PROPN bracis-28410 339 20 , , PUNCT bracis-28410 339 21 c.c c.c PROPN bracis-28410 339 22 . . PROPN bracis-28410 339 23 , , PUNCT bracis-28410 339 24 he he PRON bracis-28410 339 25 , , PUNCT bracis-28410 339 26 k. k. PROPN bracis-28410 339 27 , , PUNCT bracis-28410 339 28 tang tang PROPN bracis-28410 339 29 , , PUNCT bracis-28410 339 30 x. x. NOUN bracis-28410 339 31 : : PUNCT bracis-28410 339 32 image image NOUN bracis-28410 339 33 super super NOUN bracis-28410 339 34 - - NOUN bracis-28410 339 35 resolution resolution NOUN bracis-28410 339 36 using use VERB bracis-28410 339 37 deep deep ADJ bracis-28410 339 38 convolutional convolutional ADJ bracis-28410 339 39 networks network NOUN bracis-28410 339 40 . . PUNCT bracis-28410 340 1 trans trans PROPN bracis-28410 340 2 . . PROPN bracis-28410 340 3 pattern pattern PROPN bracis-28410 340 4 anal anal PROPN bracis-28410 340 5 . . PUNCT bracis-28410 341 1 mach mach PROPN bracis-28410 341 2 . . PUNCT bracis-28410 342 1 intell intell PROPN bracis-28410 342 2 . . PUNCT bracis-28410 343 1 38(2 38(2 NUM bracis-28410 343 2 ) ) PUNCT bracis-28410 343 3 , , PUNCT bracis-28410 343 4 295–307 295–307 NUM bracis-28410 343 5 ( ( PUNCT bracis-28410 343 6 2016 2016 NUM bracis-28410 343 7 ) ) PUNCT bracis-28410 343 8 article article NOUN bracis-28410 343 9     SPACE bracis-28410 343 10 google google PROPN bracis-28410 343 11 scholar scholar NOUN bracis-28410 343 12     SPACE bracis-28410 343 13 dong dong PROPN bracis-28410 343 14 , , PUNCT bracis-28410 343 15 c. c. PROPN bracis-28410 343 16 , , PUNCT bracis-28410 343 17 loy loy PROPN bracis-28410 343 18 , , PUNCT bracis-28410 343 19 c.c c.c PROPN bracis-28410 343 20 . . PROPN bracis-28410 343 21 , , PUNCT bracis-28410 343 22 he he PRON bracis-28410 343 23 , , PUNCT bracis-28410 343 24 k. k. PROPN bracis-28410 343 25 , , PUNCT bracis-28410 343 26 tang tang PROPN bracis-28410 343 27 , , PUNCT bracis-28410 343 28 x. x. NOUN bracis-28410 343 29 : : PUNCT bracis-28410 343 30 learning learn VERB bracis-28410 343 31 a a DET bracis-28410 343 32 deep deep ADJ bracis-28410 343 33 convolutional convolutional ADJ bracis-28410 343 34 network network NOUN bracis-28410 343 35 for for ADP bracis-28410 343 36 image image NOUN bracis-28410 343 37 super super NOUN bracis-28410 343 38 - - NOUN bracis-28410 343 39 resolution resolution NOUN bracis-28410 343 40 . . PUNCT bracis-28410 344 1 in in ADP bracis-28410 344 2 : : PUNCT bracis-28410 344 3 fleet fleet PROPN bracis-28410 344 4 , , PUNCT bracis-28410 344 5 d. d. PROPN bracis-28410 344 6 , , PUNCT bracis-28410 344 7 pajdla pajdla NOUN bracis-28410 344 8 , , PUNCT bracis-28410 344 9 t. t. PROPN bracis-28410 344 10 , , PUNCT bracis-28410 344 11 schiele schiele PROPN bracis-28410 344 12 , , PUNCT bracis-28410 344 13 b. b. PROPN bracis-28410 344 14 , , PUNCT bracis-28410 344 15 tuytelaars tuytelaar NOUN bracis-28410 344 16 , , PUNCT bracis-28410 344 17 t. t. PROPN bracis-28410 344 18 ( ( PUNCT bracis-28410 344 19 eds eds PROPN bracis-28410 344 20 . . PUNCT bracis-28410 344 21 ) ) PUNCT bracis-28410 345 1 eccv eccv ADV bracis-28410 345 2 2014 2014 NUM bracis-28410 345 3 . . PUNCT bracis-28410 346 1 lncs lncs PROPN bracis-28410 346 2 , , PUNCT bracis-28410 346 3 vol vol NOUN bracis-28410 346 4 . . PROPN bracis-28410 346 5 8692 8692 NUM bracis-28410 346 6 , , PUNCT bracis-28410 346 7 pp pp ADV bracis-28410 346 8 . . PUNCT bracis-28410 347 1 184–199 184–199 NUM bracis-28410 347 2 . . PUNCT bracis-28410 347 3 springer springer NOUN bracis-28410 347 4 , , PUNCT bracis-28410 347 5 cham cham PROPN bracis-28410 347 6 ( ( PUNCT bracis-28410 347 7 2014 2014 NUM bracis-28410 347 8 ) ) PUNCT bracis-28410 347 9 . . PUNCT bracis-28410 348 1 https://doi.org/10.1007/978-3-319-10593-2_13 https://doi.org/10.1007/978-3-319-10593-2_13 PROPN bracis-28410 348 2 chapter chapter NOUN bracis-28410 348 3     SPACE bracis-28410 348 4 google google PROPN bracis-28410 348 5 scholar scholar NOUN bracis-28410 348 6     SPACE bracis-28410 348 7 dong dong PROPN bracis-28410 348 8 , , PUNCT bracis-28410 348 9 c. c. PROPN bracis-28410 348 10 , , PUNCT bracis-28410 348 11 loy loy PROPN bracis-28410 348 12 , , PUNCT bracis-28410 348 13 c.c c.c PROPN bracis-28410 348 14 . . PROPN bracis-28410 348 15 , , PUNCT bracis-28410 348 16 tang tang PROPN bracis-28410 348 17 , , PUNCT bracis-28410 348 18 x. x. NOUN bracis-28410 348 19 : : PUNCT bracis-28410 348 20 accelerating accelerate VERB bracis-28410 348 21 the the DET bracis-28410 348 22 super super ADJ bracis-28410 348 23 - - ADJ bracis-28410 348 24 resolution resolution ADJ bracis-28410 348 25 convolutional convolutional ADJ bracis-28410 348 26 neural neural ADJ bracis-28410 348 27 network network NOUN bracis-28410 348 28 . . PUNCT bracis-28410 349 1 in in ADP bracis-28410 349 2 : : PUNCT bracis-28410 349 3 leibe leibe ADJ bracis-28410 349 4 , , PUNCT bracis-28410 349 5 b. b. PROPN bracis-28410 349 6 , , PUNCT bracis-28410 349 7 matas matas PROPN bracis-28410 349 8 , , PUNCT bracis-28410 349 9 j. j. PROPN bracis-28410 349 10 , , PUNCT bracis-28410 349 11 sebe sebe PROPN bracis-28410 349 12 , , PUNCT bracis-28410 349 13 n. n. NOUN bracis-28410 349 14 , , PUNCT bracis-28410 349 15 welling well VERB bracis-28410 349 16 , , PUNCT bracis-28410 349 17 m. m. NOUN bracis-28410 349 18 ( ( PUNCT bracis-28410 349 19 eds ed NOUN bracis-28410 349 20 . . PUNCT bracis-28410 349 21 ) ) PUNCT bracis-28410 349 22 eccv eccv ADV bracis-28410 349 23 2016 2016 NUM bracis-28410 349 24 . . PUNCT bracis-28410 350 1 lncs lncs PROPN bracis-28410 350 2 , , PUNCT bracis-28410 350 3 vol vol NOUN bracis-28410 350 4 . . PROPN bracis-28410 350 5 9906 9906 NUM bracis-28410 350 6 , , PUNCT bracis-28410 350 7 pp pp ADJ bracis-28410 350 8 . . PUNCT bracis-28410 351 1 391–407 391–407 NUM bracis-28410 351 2 . . X bracis-28410 351 3 springer springer NOUN bracis-28410 351 4 , , PUNCT bracis-28410 351 5 cham cham PROPN bracis-28410 351 6 ( ( PUNCT bracis-28410 351 7 2016 2016 NUM bracis-28410 351 8 ) ) PUNCT bracis-28410 351 9 . . PUNCT bracis-28410 352 1 https://doi.org/10.1007/978-3-319-46475-6_25 https://doi.org/10.1007/978-3-319-46475-6_25 PROPN bracis-28410 352 2 chapter chapter NOUN bracis-28410 352 3     SPACE bracis-28410 352 4 google google PROPN bracis-28410 352 5 scholar scholar NOUN bracis-28410 352 6     SPACE bracis-28410 352 7 gu gu NOUN bracis-28410 352 8 , , PUNCT bracis-28410 352 9 j. j. PROPN bracis-28410 352 10 , , PUNCT bracis-28410 352 11 tresp tresp NOUN bracis-28410 352 12 , , PUNCT bracis-28410 352 13 v. v. ADV bracis-28410 352 14 : : PUNCT bracis-28410 352 15 improving improve VERB bracis-28410 352 16 the the DET bracis-28410 352 17 robustness robustness NOUN bracis-28410 352 18 of of ADP bracis-28410 352 19 capsule capsule NOUN bracis-28410 352 20 networks network NOUN bracis-28410 352 21 to to ADP bracis-28410 352 22 image image NOUN bracis-28410 352 23 affine affine NOUN bracis-28410 352 24 transformations transformation NOUN bracis-28410 352 25 . . PUNCT bracis-28410 353 1 in in ADP bracis-28410 353 2 : : PUNCT bracis-28410 353 3 cvpr cvpr NOUN bracis-28410 353 4 , , PUNCT bracis-28410 353 5 pp pp CCONJ bracis-28410 353 6 . . PUNCT bracis-28410 354 1 1–15 1–15 NUM bracis-28410 354 2 ( ( PUNCT bracis-28410 354 3 2020 2020 NUM bracis-28410 354 4 ) ) PUNCT bracis-28410 354 5 google google PROPN bracis-28410 354 6 scholar scholar NOUN bracis-28410 354 7     SPACE bracis-28410 354 8 gu gu PROPN bracis-28410 354 9 , , PUNCT bracis-28410 354 10 j. j. PROPN bracis-28410 354 11 , , PUNCT bracis-28410 354 12 wu wu PROPN bracis-28410 354 13 , , PUNCT bracis-28410 354 14 b. b. PROPN bracis-28410 354 15 , , PUNCT bracis-28410 354 16 tresp tresp NOUN bracis-28410 354 17 , , PUNCT bracis-28410 354 18 v. v. CCONJ bracis-28410 354 19 : : PUNCT bracis-28410 354 20 effective effective ADJ bracis-28410 354 21 and and CCONJ bracis-28410 354 22 efficient efficient ADJ bracis-28410 354 23 vote vote NOUN bracis-28410 354 24 atack atack NOUN bracis-28410 354 25 on on ADP bracis-28410 354 26 capsule capsule NOUN bracis-28410 354 27 networks network NOUN bracis-28410 354 28 . . PUNCT bracis-28410 355 1 in in ADP bracis-28410 355 2 : : PUNCT bracis-28410 355 3 iclr iclr NOUN bracis-28410 355 4 ( ( PUNCT bracis-28410 355 5 2021 2021 NUM bracis-28410 355 6 ) ) PUNCT bracis-28410 355 7 google google PROPN bracis-28410 355 8 scholar scholar NOUN bracis-28410 355 9     SPACE bracis-28410 355 10 hinton hinton PROPN bracis-28410 355 11 , , PUNCT bracis-28410 355 12 g. g. PROPN bracis-28410 355 13 , , PUNCT bracis-28410 355 14 sabour sabour NOUN bracis-28410 355 15 , , PUNCT bracis-28410 355 16 s. s. PROPN bracis-28410 355 17 , , PUNCT bracis-28410 355 18 frosst frosst ADJ bracis-28410 355 19 , , PUNCT bracis-28410 355 20 n. n. NOUN bracis-28410 355 21 : : PUNCT bracis-28410 355 22 matrix matrix NOUN bracis-28410 355 23 capsules capsule NOUN bracis-28410 355 24 with with ADP bracis-28410 355 25 em em PRON bracis-28410 355 26 routing route VERB bracis-28410 355 27 . . PUNCT bracis-28410 356 1 in in ADP bracis-28410 356 2 : : PUNCT bracis-28410 356 3 iclr iclr NOUN bracis-28410 356 4 , , PUNCT bracis-28410 356 5 pp pp ADP bracis-28410 356 6 . . PUNCT bracis-28410 357 1 1–10 1–10 NOUN bracis-28410 357 2 ( ( PUNCT bracis-28410 357 3 2018 2018 NUM bracis-28410 357 4 ) ) PUNCT bracis-28410 357 5 google google PROPN bracis-28410 357 6 scholar scholar PROPN bracis-28410 357 7     SPACE bracis-28410 357 8 hinton hinton PROPN bracis-28410 357 9 , , PUNCT bracis-28410 357 10 g.e g.e PROPN bracis-28410 357 11 . . PROPN bracis-28410 357 12 , , PUNCT bracis-28410 357 13 krizhevsky krizhevsky PROPN bracis-28410 357 14 , , PUNCT bracis-28410 357 15 a. a. PROPN bracis-28410 357 16 , , PUNCT bracis-28410 357 17 wang wang PROPN bracis-28410 357 18 , , PUNCT bracis-28410 357 19 s.d s.d PROPN bracis-28410 357 20 . . PROPN bracis-28410 357 21 : : PUNCT bracis-28410 357 22 transforming transform VERB bracis-28410 357 23 auto auto NOUN bracis-28410 357 24 - - PUNCT bracis-28410 357 25 encoders encoder NOUN bracis-28410 357 26 . . PUNCT bracis-28410 358 1 in in ADP bracis-28410 358 2 : : PUNCT bracis-28410 358 3 artificial artificial ADJ bracis-28410 358 4 neural neural ADJ bracis-28410 358 5 networks network NOUN bracis-28410 358 6 and and CCONJ bracis-28410 358 7 machine machine NOUN bracis-28410 358 8 learning learning NOUN bracis-28410 358 9 , , PUNCT bracis-28410 358 10 pp pp ADP bracis-28410 358 11 . . PUNCT bracis-28410 359 1 44–51 44–51 NUM bracis-28410 359 2 ( ( PUNCT bracis-28410 359 3 2011 2011 NUM bracis-28410 359 4 ) ) PUNCT bracis-28410 359 5 google google PROPN bracis-28410 359 6 scholar scholar NOUN bracis-28410 359 7     SPACE bracis-28410 359 8 hore hore ADJ bracis-28410 359 9 , , PUNCT bracis-28410 359 10 a. a. NOUN bracis-28410 359 11 , , PUNCT bracis-28410 359 12 ziou ziou NOUN bracis-28410 359 13 , , PUNCT bracis-28410 359 14 d. d. PROPN bracis-28410 359 15 : : PUNCT bracis-28410 359 16 image image NOUN bracis-28410 359 17 quality quality NOUN bracis-28410 359 18 metrics metric NOUN bracis-28410 359 19 : : PUNCT bracis-28410 359 20 psnr psnr NOUN bracis-28410 359 21 vs. vs. ADP bracis-28410 359 22 ssim ssim NOUN bracis-28410 359 23 . . PUNCT bracis-28410 360 1 in in ADP bracis-28410 360 2 : : PUNCT bracis-28410 360 3 icpr icpr PROPN bracis-28410 360 4 , , PUNCT bracis-28410 360 5 pp pp X bracis-28410 360 6 . . PUNCT bracis-28410 361 1 2366–2369 2366–2369 NUM bracis-28410 361 2 ( ( PUNCT bracis-28410 361 3 2010 2010 NUM bracis-28410 361 4 ) ) PUNCT bracis-28410 361 5 google google PROPN bracis-28410 361 6 scholar scholar NOUN bracis-28410 361 7     SPACE bracis-28410 361 8 hsu hsu PROPN bracis-28410 361 9 , , PUNCT bracis-28410 361 10 j. j. PROPN bracis-28410 361 11 , , PUNCT bracis-28410 361 12 kuo kuo PROPN bracis-28410 361 13 , , PUNCT bracis-28410 361 14 c. c. PROPN bracis-28410 361 15 , , PUNCT bracis-28410 361 16 chen chen PROPN bracis-28410 361 17 , , PUNCT bracis-28410 361 18 d. d. PROPN bracis-28410 361 19 : : PUNCT bracis-28410 361 20 image image VERB bracis-28410 361 21 super super NOUN bracis-28410 361 22 - - NOUN bracis-28410 361 23 resolution resolution NOUN bracis-28410 361 24 using use VERB bracis-28410 361 25 capsule capsule ADJ bracis-28410 361 26 neural neural ADJ bracis-28410 361 27 networks network NOUN bracis-28410 361 28 . . PUNCT bracis-28410 362 1 ieee ieee NOUN bracis-28410 362 2 access access NOUN bracis-28410 362 3 ( ( PUNCT bracis-28410 362 4 2020 2020 NUM bracis-28410 362 5 ) ) PUNCT bracis-28410 362 6 google google PROPN bracis-28410 362 7 scholar scholar NOUN bracis-28410 362 8     SPACE bracis-28410 362 9 huang huang PROPN bracis-28410 362 10 , , PUNCT bracis-28410 362 11 j.b j.b PROPN bracis-28410 362 12 . . PROPN bracis-28410 362 13 , , PUNCT bracis-28410 362 14 singh singh PROPN bracis-28410 362 15 , , PUNCT bracis-28410 362 16 a. a. PROPN bracis-28410 362 17 , , PUNCT bracis-28410 362 18 ahuja ahuja PROPN bracis-28410 362 19 , , PUNCT bracis-28410 362 20 n. n. NOUN bracis-28410 362 21 : : PUNCT bracis-28410 362 22 single single ADJ bracis-28410 362 23 image image NOUN bracis-28410 362 24 super super NOUN bracis-28410 362 25 - - NOUN bracis-28410 362 26 resolution resolution NOUN bracis-28410 362 27 from from ADP bracis-28410 362 28 transformed transform VERB bracis-28410 362 29 self self NOUN bracis-28410 362 30 - - PUNCT bracis-28410 362 31 exemplars exemplar NOUN bracis-28410 362 32 . . PUNCT bracis-28410 363 1 in in ADP bracis-28410 363 2 : : PUNCT bracis-28410 363 3 cvpr cvpr NOUN bracis-28410 363 4 , , PUNCT bracis-28410 363 5 pp pp ADP bracis-28410 363 6 . . PUNCT bracis-28410 364 1 1–9 1–9 NUM bracis-28410 364 2 ( ( PUNCT bracis-28410 364 3 2015 2015 NUM bracis-28410 364 4 ) ) PUNCT bracis-28410 364 5 google google PROPN bracis-28410 364 6 scholar scholar NOUN bracis-28410 364 7     SPACE bracis-28410 364 8 huang huang PROPN bracis-28410 364 9 , , PUNCT bracis-28410 364 10 w. w. PROPN bracis-28410 364 11 , , PUNCT bracis-28410 364 12 zhou zhou PROPN bracis-28410 364 13 , , PUNCT bracis-28410 364 14 f. f. PROPN bracis-28410 364 15 : : PUNCT bracis-28410 364 16 da da ADJ bracis-28410 364 17 - - ADJ bracis-28410 364 18 capsnet capsnet NOUN bracis-28410 364 19 : : PUNCT bracis-28410 364 20 dual dual ADJ bracis-28410 364 21 attention attention NOUN bracis-28410 364 22 mechanism mechanism NOUN bracis-28410 364 23 capsule capsule NOUN bracis-28410 364 24 network network NOUN bracis-28410 364 25 . . PUNCT bracis-28410 365 1 sci sci PROPN bracis-28410 365 2 . . PROPN bracis-28410 365 3 rep rep AUX bracis-28410 365 4 . . PROPN bracis-28410 365 5 ( ( PUNCT bracis-28410 365 6 2020 2020 NUM bracis-28410 365 7 ) ) PUNCT bracis-28410 365 8 google google PROPN bracis-28410 365 9 scholar scholar NOUN bracis-28410 365 10     SPACE bracis-28410 365 11 irani irani NOUN bracis-28410 365 12 , , PUNCT bracis-28410 365 13 m. m. NOUN bracis-28410 365 14 , , PUNCT bracis-28410 365 15 peleg peleg NOUN bracis-28410 365 16 , , PUNCT bracis-28410 365 17 s. s. PROPN bracis-28410 365 18 : : PUNCT bracis-28410 365 19 improving improve VERB bracis-28410 365 20 resolution resolution NOUN bracis-28410 365 21 by by ADP bracis-28410 365 22 image image NOUN bracis-28410 365 23 registration registration NOUN bracis-28410 365 24 . . PUNCT bracis-28410 366 1 in in ADP bracis-28410 366 2 : : PUNCT bracis-28410 366 3 cvgip cvgip ADJ bracis-28410 366 4 : : PUNCT bracis-28410 366 5 graph graph NOUN bracis-28410 366 6 . . PUNCT bracis-28410 366 7 model model NOUN bracis-28410 366 8 . . PUNCT bracis-28410 366 9 image image NOUN bracis-28410 366 10 process process NOUN bracis-28410 366 11 . . PUNCT bracis-28410 367 1 53(3 53(3 NUM bracis-28410 367 2 ) ) PUNCT bracis-28410 367 3 , , PUNCT bracis-28410 367 4 231–239 231–239 NUM bracis-28410 367 5 ( ( PUNCT bracis-28410 367 6 1991 1991 NUM bracis-28410 367 7 ) ) PUNCT bracis-28410 367 8 google google PROPN bracis-28410 367 9 scholar scholar NOUN bracis-28410 367 10     SPACE bracis-28410 367 11 ji ji PROPN bracis-28410 367 12 , , PUNCT bracis-28410 367 13 x. x. PROPN bracis-28410 367 14 , , PUNCT bracis-28410 367 15 cao cao PROPN bracis-28410 367 16 , , PUNCT bracis-28410 367 17 y. y. PROPN bracis-28410 367 18 , , PUNCT bracis-28410 367 19 tai tai PROPN bracis-28410 367 20 , , PUNCT bracis-28410 367 21 y. y. PROPN bracis-28410 367 22 , , PUNCT bracis-28410 367 23 wang wang PROPN bracis-28410 367 24 , , PUNCT bracis-28410 367 25 c. c. PROPN bracis-28410 367 26 , , PUNCT bracis-28410 367 27 li li PROPN bracis-28410 367 28 , , PUNCT bracis-28410 367 29 j. j. PROPN bracis-28410 367 30 , , PUNCT bracis-28410 367 31 huang huang PROPN bracis-28410 367 32 , , PUNCT bracis-28410 367 33 f. f. PROPN bracis-28410 367 34 : : PUNCT bracis-28410 367 35 real real ADJ bracis-28410 367 36 - - PUNCT bracis-28410 367 37 world world NOUN bracis-28410 367 38 super super NOUN bracis-28410 367 39 - - NOUN bracis-28410 367 40 resolution resolution NOUN bracis-28410 367 41 via via ADP bracis-28410 367 42 kernel kernel PROPN bracis-28410 367 43 estimation estimation NOUN bracis-28410 367 44 and and CCONJ bracis-28410 367 45 noise noise NOUN bracis-28410 367 46 injection injection NOUN bracis-28410 367 47 . . PUNCT bracis-28410 368 1 in in ADP bracis-28410 368 2 : : PUNCT bracis-28410 368 3 cvpr cvpr NOUN bracis-28410 368 4 workshops workshop NOUN bracis-28410 368 5 , , PUNCT bracis-28410 368 6 pp pp ADV bracis-28410 368 7 . . PUNCT bracis-28410 369 1 1–8 1–8 NUM bracis-28410 369 2 ( ( PUNCT bracis-28410 369 3 2020 2020 NUM bracis-28410 369 4 ) ) PUNCT bracis-28410 369 5 google google PROPN bracis-28410 369 6 scholar scholar NOUN bracis-28410 369 7     SPACE bracis-28410 369 8 kastryulin kastryulin PROPN bracis-28410 369 9 , , PUNCT bracis-28410 369 10 s. s. PROPN bracis-28410 369 11 , , PUNCT bracis-28410 369 12 zakirov zakirov PROPN bracis-28410 369 13 , , PUNCT bracis-28410 369 14 d. d. PROPN bracis-28410 369 15 , , PUNCT bracis-28410 369 16 prokopenko prokopenko PROPN bracis-28410 369 17 , , PUNCT bracis-28410 369 18 d. d. PROPN bracis-28410 369 19 : : PUNCT bracis-28410 369 20 pytorch pytorch NOUN bracis-28410 369 21 image image NOUN bracis-28410 369 22 quality quality NOUN bracis-28410 369 23 : : PUNCT bracis-28410 369 24 metrics metric NOUN bracis-28410 369 25 and and CCONJ bracis-28410 369 26 measure measure NOUN bracis-28410 369 27 for for ADP bracis-28410 369 28 image image NOUN bracis-28410 369 29 quality quality NOUN bracis-28410 369 30 assessment assessment NOUN bracis-28410 369 31 ( ( PUNCT bracis-28410 369 32 2019 2019 NUM bracis-28410 369 33 ) ) PUNCT bracis-28410 369 34 . . PUNCT bracis-28410 370 1 https://github.com/photosynthesis-team/piq https://github.com/photosynthesis-team/piq PROPN bracis-28410 370 2 kim kim PROPN bracis-28410 370 3 , , PUNCT bracis-28410 370 4 j. j. PROPN bracis-28410 370 5 , , PUNCT bracis-28410 370 6 kwon kwon PROPN bracis-28410 370 7 lee lee PROPN bracis-28410 370 8 , , PUNCT bracis-28410 370 9 j. j. PROPN bracis-28410 370 10 , , PUNCT bracis-28410 370 11 mu mu PROPN bracis-28410 370 12 lee lee PROPN bracis-28410 370 13 , , PUNCT bracis-28410 370 14 k. k. PROPN bracis-28410 370 15 : : PUNCT bracis-28410 370 16 accurate accurate ADJ bracis-28410 370 17 image image NOUN bracis-28410 370 18 super super NOUN bracis-28410 370 19 - - NOUN bracis-28410 370 20 resolution resolution NOUN bracis-28410 370 21 using use VERB bracis-28410 370 22 very very ADV bracis-28410 370 23 deep deep ADJ bracis-28410 370 24 convolutional convolutional ADJ bracis-28410 370 25 networks network NOUN bracis-28410 370 26 . . PUNCT bracis-28410 371 1 in in ADP bracis-28410 371 2 : : PUNCT bracis-28410 371 3 cvpr cvpr NOUN bracis-28410 371 4 , , PUNCT bracis-28410 371 5 pp pp ADV bracis-28410 371 6 . . PUNCT bracis-28410 372 1 1–8 1–8 NUM bracis-28410 372 2 ( ( PUNCT bracis-28410 372 3 2016 2016 NUM bracis-28410 372 4 ) ) PUNCT bracis-28410 372 5 google google PROPN bracis-28410 372 6 scholar scholar NOUN bracis-28410 372 7     SPACE bracis-28410 372 8 kim kim PROPN bracis-28410 372 9 , , PUNCT bracis-28410 372 10 j. j. PROPN bracis-28410 372 11 , , PUNCT bracis-28410 372 12 kwon kwon PROPN bracis-28410 372 13 lee lee PROPN bracis-28410 372 14 , , PUNCT bracis-28410 372 15 j. j. PROPN bracis-28410 372 16 , , PUNCT bracis-28410 372 17 mu mu PROPN bracis-28410 372 18 lee lee PROPN bracis-28410 372 19 , , PUNCT bracis-28410 372 20 k. k. PROPN bracis-28410 372 21 : : PUNCT bracis-28410 372 22 deeply deeply ADV bracis-28410 372 23 - - PUNCT bracis-28410 372 24 recursive recursive ADJ bracis-28410 372 25 convolutional convolutional ADJ bracis-28410 372 26 network network NOUN bracis-28410 372 27 for for ADP bracis-28410 372 28 image image NOUN bracis-28410 372 29 super super NOUN bracis-28410 372 30 - - NOUN bracis-28410 372 31 resolution resolution NOUN bracis-28410 372 32 . . PUNCT bracis-28410 373 1 in in ADP bracis-28410 373 2 : : PUNCT bracis-28410 373 3 cvpr cvpr NOUN bracis-28410 373 4 , , PUNCT bracis-28410 373 5 pp pp ADP bracis-28410 373 6 . . PUNCT bracis-28410 374 1 1–13 1–13 NOUN bracis-28410 374 2 ( ( PUNCT bracis-28410 374 3 2016 2016 NUM bracis-28410 374 4 ) ) PUNCT bracis-28410 374 5 google google PROPN bracis-28410 374 6 scholar scholar NOUN bracis-28410 374 7     SPACE bracis-28410 374 8 kim kim PROPN bracis-28410 374 9 , , PUNCT bracis-28410 374 10 j.h j.h PROPN bracis-28410 374 11 . . PROPN bracis-28410 374 12 , , PUNCT bracis-28410 374 13 lee lee PROPN bracis-28410 374 14 , , PUNCT bracis-28410 374 15 j.s j.s PROPN bracis-28410 374 16 . . PUNCT bracis-28410 374 17 : : PUNCT bracis-28410 375 1 deep deep ADJ bracis-28410 375 2 residual residual ADJ bracis-28410 375 3 network network NOUN bracis-28410 375 4 with with ADP bracis-28410 375 5 enhanced enhanced ADJ bracis-28410 375 6 upscaling upscaling NOUN bracis-28410 375 7 module module NOUN bracis-28410 375 8 for for ADP bracis-28410 375 9 super super NOUN bracis-28410 375 10 - - NOUN bracis-28410 375 11 resolution resolution NOUN bracis-28410 375 12 . . PUNCT bracis-28410 376 1 in in ADP bracis-28410 376 2 : : PUNCT bracis-28410 376 3 cvpr cvpr NOUN bracis-28410 376 4 workshops workshop NOUN bracis-28410 376 5 , , PUNCT bracis-28410 376 6 pp pp ADV bracis-28410 376 7 . . PUNCT bracis-28410 377 1 1–15 1–15 NUM bracis-28410 377 2 ( ( PUNCT bracis-28410 377 3 2018 2018 NUM bracis-28410 377 4 ) ) PUNCT bracis-28410 377 5 google google PROPN bracis-28410 377 6 scholar scholar NOUN bracis-28410 377 7     SPACE bracis-28410 377 8 kingma kingma PROPN bracis-28410 377 9 , , PUNCT bracis-28410 377 10 d.p d.p PROPN bracis-28410 377 11 . . PROPN bracis-28410 377 12 , , PUNCT bracis-28410 377 13 ba ba PROPN bracis-28410 377 14 , , PUNCT bracis-28410 377 15 j. j. PROPN bracis-28410 377 16 : : PUNCT bracis-28410 377 17 adam adam PROPN bracis-28410 377 18 : : PUNCT bracis-28410 377 19 a a DET bracis-28410 377 20 method method NOUN bracis-28410 377 21 for for ADP bracis-28410 377 22 stochastic stochastic ADJ bracis-28410 377 23 optimization optimization NOUN bracis-28410 377 24 . . PUNCT bracis-28410 378 1 in in ADP bracis-28410 378 2 : : PUNCT bracis-28410 378 3 bengio bengio PROPN bracis-28410 378 4 , , PUNCT bracis-28410 378 5 y. y. PROPN bracis-28410 378 6 , , PUNCT bracis-28410 378 7 lecun lecun ADJ bracis-28410 378 8 , , PUNCT bracis-28410 378 9 y. y. PROPN bracis-28410 378 10 ( ( PUNCT bracis-28410 378 11 eds eds PROPN bracis-28410 378 12 . . PUNCT bracis-28410 378 13 ) ) PUNCT bracis-28410 379 1 iclr iclr NOUN bracis-28410 379 2 ( ( PUNCT bracis-28410 379 3 2015 2015 NUM bracis-28410 379 4 ) ) PUNCT bracis-28410 379 5 google google PROPN bracis-28410 379 6 scholar scholar NOUN bracis-28410 379 7     SPACE bracis-28410 379 8 lalonde lalonde NOUN bracis-28410 379 9 , , PUNCT bracis-28410 379 10 r. r. PROPN bracis-28410 379 11 , , PUNCT bracis-28410 379 12 bagci bagci PROPN bracis-28410 379 13 , , PUNCT bracis-28410 379 14 u. u. NOUN bracis-28410 379 15 : : PUNCT bracis-28410 379 16 capsules capsule NOUN bracis-28410 379 17 for for ADP bracis-28410 379 18 object object NOUN bracis-28410 379 19 segmentation segmentation NOUN bracis-28410 379 20 . . PUNCT bracis-28410 380 1 arxiv arxiv PROPN bracis-28410 380 2 preprint preprint VERB bracis-28410 380 3 arxiv:1804.04241 arxiv:1804.04241 NUM bracis-28410 380 4 ( ( PUNCT bracis-28410 380 5 2018 2018 NUM bracis-28410 380 6 ) ) PUNCT bracis-28410 380 7 lecun lecun ADJ bracis-28410 380 8 , , PUNCT bracis-28410 380 9 y. y. PROPN bracis-28410 380 10 , , PUNCT bracis-28410 380 11 bottou bottou PROPN bracis-28410 380 12 , , PUNCT bracis-28410 380 13 l. l. PROPN bracis-28410 380 14 , , PUNCT bracis-28410 380 15 bengio bengio PROPN bracis-28410 380 16 , , PUNCT bracis-28410 380 17 y. y. PROPN bracis-28410 380 18 , , PUNCT bracis-28410 380 19 haffner haffner NOUN bracis-28410 380 20 , , PUNCT bracis-28410 380 21 p. p. NOUN bracis-28410 380 22 : : PUNCT bracis-28410 381 1 gradient gradient NOUN bracis-28410 381 2 - - PUNCT bracis-28410 381 3 based base VERB bracis-28410 381 4 learning learning NOUN bracis-28410 381 5 applied apply VERB bracis-28410 381 6 to to ADP bracis-28410 381 7 document document NOUN bracis-28410 381 8 recognition recognition NOUN bracis-28410 381 9 . . PUNCT bracis-28410 382 1 proc proc PROPN bracis-28410 382 2 . . PUNCT bracis-28410 383 1 ieee ieee PROPN bracis-28410 383 2 86(11 86(11 NUM bracis-28410 383 3 ) ) PUNCT bracis-28410 383 4 , , PUNCT bracis-28410 383 5 2278–2324 2278–2324 NUM bracis-28410 383 6 ( ( PUNCT bracis-28410 383 7 1998 1998 NUM bracis-28410 383 8 ) ) PUNCT bracis-28410 383 9 article article NOUN bracis-28410 383 10     SPACE bracis-28410 383 11 google google PROPN bracis-28410 383 12 scholar scholar NOUN bracis-28410 383 13     SPACE bracis-28410 383 14 ledig ledig NOUN bracis-28410 383 15 , , PUNCT bracis-28410 383 16 c. c. PROPN bracis-28410 383 17 , , PUNCT bracis-28410 383 18 et et PROPN bracis-28410 383 19 al al PROPN bracis-28410 383 20 . . PROPN bracis-28410 383 21 : : PUNCT bracis-28410 384 1 photo photo ADJ bracis-28410 384 2 - - PUNCT bracis-28410 384 3 realistic realistic ADJ bracis-28410 384 4 single single ADJ bracis-28410 384 5 image image NOUN bracis-28410 384 6 super super NOUN bracis-28410 384 7 - - NOUN bracis-28410 384 8 resolution resolution NOUN bracis-28410 384 9 using use VERB bracis-28410 384 10 a a DET bracis-28410 384 11 generative generative ADJ bracis-28410 384 12 adversarial adversarial ADJ bracis-28410 384 13 network network NOUN bracis-28410 384 14 . . PUNCT bracis-28410 385 1 in in ADP bracis-28410 385 2 : : PUNCT bracis-28410 385 3 cvpr cvpr NOUN bracis-28410 385 4 , , PUNCT bracis-28410 385 5 pp pp ADV bracis-28410 385 6 . . PUNCT bracis-28410 386 1 1–8 1–8 NUM bracis-28410 386 2 ( ( PUNCT bracis-28410 386 3 2017 2017 NUM bracis-28410 386 4 ) ) PUNCT bracis-28410 386 5 google google PROPN bracis-28410 386 6 scholar scholar NOUN bracis-28410 386 7     SPACE bracis-28410 386 8 li li PROPN bracis-28410 386 9 , , PUNCT bracis-28410 386 10 c. c. NOUN bracis-28410 386 11 , , PUNCT bracis-28410 386 12 bovik bovik NOUN bracis-28410 386 13 , , PUNCT bracis-28410 386 14 a.c a.c PROPN bracis-28410 386 15 . . PROPN bracis-28410 386 16 : : PUNCT bracis-28410 387 1 content content NOUN bracis-28410 387 2 - - PUNCT bracis-28410 387 3 weighted weight VERB bracis-28410 387 4 video video NOUN bracis-28410 387 5 quality quality NOUN bracis-28410 387 6 assessment assessment NOUN bracis-28410 387 7 using use VERB bracis-28410 387 8 a a DET bracis-28410 387 9 three three NUM bracis-28410 387 10 - - PUNCT bracis-28410 387 11 component component NOUN bracis-28410 387 12 image image NOUN bracis-28410 387 13 model model NOUN bracis-28410 387 14 . . PUNCT bracis-28410 388 1 j. j. PROPN bracis-28410 388 2 electron electron PROPN bracis-28410 388 3 . . PUNCT bracis-28410 389 1 imag imag PROPN bracis-28410 389 2 . . PROPN bracis-28410 390 1 19 19 NUM bracis-28410 390 2 , , PUNCT bracis-28410 390 3 19 19 NUM bracis-28410 390 4 ( ( PUNCT bracis-28410 390 5 2010 2010 NUM bracis-28410 390 6 ) ) PUNCT bracis-28410 390 7 google google PROPN bracis-28410 390 8 scholar scholar NOUN bracis-28410 390 9     SPACE bracis-28410 390 10 liaw liaw PROPN bracis-28410 390 11 , , PUNCT bracis-28410 390 12 r. r. PROPN bracis-28410 390 13 , , PUNCT bracis-28410 390 14 liang liang PROPN bracis-28410 390 15 , , PUNCT bracis-28410 390 16 e. e. PROPN bracis-28410 390 17 , , PUNCT bracis-28410 390 18 nishihara nishihara PROPN bracis-28410 390 19 , , PUNCT bracis-28410 390 20 r. r. PROPN bracis-28410 390 21 , , PUNCT bracis-28410 390 22 moritz moritz PROPN bracis-28410 390 23 , , PUNCT bracis-28410 390 24 p. p. PROPN bracis-28410 390 25 , , PUNCT bracis-28410 390 26 gonzalez gonzalez PROPN bracis-28410 390 27 , , PUNCT bracis-28410 390 28 j.e j.e PROPN bracis-28410 390 29 . . PROPN bracis-28410 390 30 , , PUNCT bracis-28410 390 31 stoica stoica PROPN bracis-28410 390 32 , , PUNCT bracis-28410 390 33 i. i. NOUN bracis-28410 390 34 : : PUNCT bracis-28410 390 35 tune tune NOUN bracis-28410 390 36 : : PUNCT bracis-28410 390 37 a a DET bracis-28410 390 38 research research NOUN bracis-28410 390 39 platform platform NOUN bracis-28410 390 40 for for ADP bracis-28410 390 41 distributed distribute VERB bracis-28410 390 42 model model NOUN bracis-28410 390 43 selection selection NOUN bracis-28410 390 44 and and CCONJ bracis-28410 390 45 training training NOUN bracis-28410 390 46 . . PUNCT bracis-28410 391 1 arxiv arxiv PROPN bracis-28410 391 2 preprint preprint VERB bracis-28410 391 3 arxiv:1807.05118 arxiv:1807.05118 PROPN bracis-28410 391 4 ( ( PUNCT bracis-28410 391 5 2018 2018 NUM bracis-28410 391 6 ) ) PUNCT bracis-28410 392 1 lim lim PROPN bracis-28410 392 2 , , PUNCT bracis-28410 392 3 b. b. PROPN bracis-28410 392 4 , , PUNCT bracis-28410 392 5 son son PROPN bracis-28410 392 6 , , PUNCT bracis-28410 392 7 s. s. PROPN bracis-28410 392 8 , , PUNCT bracis-28410 392 9 kim kim PROPN bracis-28410 392 10 , , PUNCT bracis-28410 392 11 h. h. PROPN bracis-28410 392 12 , , PUNCT bracis-28410 392 13 nah nah INTJ bracis-28410 392 14 , , PUNCT bracis-28410 392 15 s. s. PROPN bracis-28410 392 16 , , PUNCT bracis-28410 392 17 lee lee PROPN bracis-28410 392 18 , , PUNCT bracis-28410 392 19 k.m k.m PROPN bracis-28410 392 20 . . PROPN bracis-28410 392 21 : : PUNCT bracis-28410 392 22 enhanced enhance VERB bracis-28410 392 23 deep deep ADJ bracis-28410 392 24 residual residual ADJ bracis-28410 392 25 networks network NOUN bracis-28410 392 26 for for ADP bracis-28410 392 27 single single ADJ bracis-28410 392 28 image image NOUN bracis-28410 392 29 super super NOUN bracis-28410 392 30 - - NOUN bracis-28410 392 31 resolution resolution NOUN bracis-28410 392 32 . . PUNCT bracis-28410 393 1 in in ADP bracis-28410 393 2 : : PUNCT bracis-28410 393 3 cvpr cvpr NOUN bracis-28410 393 4 workshops workshop NOUN bracis-28410 393 5 , , PUNCT bracis-28410 393 6 pp pp ADV bracis-28410 393 7 . . PUNCT bracis-28410 394 1 1–8 1–8 NUM bracis-28410 394 2 ( ( PUNCT bracis-28410 394 3 2017 2017 NUM bracis-28410 394 4 ) ) PUNCT bracis-28410 394 5 google google PROPN bracis-28410 394 6 scholar scholar NOUN bracis-28410 394 7     SPACE bracis-28410 394 8 lin lin PROPN bracis-28410 394 9 , , PUNCT bracis-28410 394 10 m. m. NOUN bracis-28410 394 11 , , PUNCT bracis-28410 394 12 chen chen PROPN bracis-28410 394 13 , , PUNCT bracis-28410 394 14 q. q. PROPN bracis-28410 394 15 , , PUNCT bracis-28410 394 16 yan yan PROPN bracis-28410 394 17 , , PUNCT bracis-28410 394 18 s. s. PROPN bracis-28410 394 19 : : PUNCT bracis-28410 394 20 network network NOUN bracis-28410 394 21 in in ADP bracis-28410 394 22 network network NOUN bracis-28410 394 23 . . PUNCT bracis-28410 395 1 arxiv arxiv PROPN bracis-28410 395 2 preprint preprint NOUN bracis-28410 395 3 arxiv:1312.4400 arxiv:1312.4400 NOUN bracis-28410 395 4 ( ( PUNCT bracis-28410 395 5 2013 2013 NUM bracis-28410 395 6 ) ) PUNCT bracis-28410 395 7 majdabadi majdabadi NOUN bracis-28410 395 8 , , PUNCT bracis-28410 395 9 m.m m.m PROPN bracis-28410 395 10 . . PROPN bracis-28410 395 11 , , PUNCT bracis-28410 395 12 ko ko PROPN bracis-28410 395 13 , , PUNCT bracis-28410 395 14 s.b s.b PROPN bracis-28410 395 15 . . PROPN bracis-28410 395 16 : : PUNCT bracis-28410 396 1 capsule capsule PROPN bracis-28410 396 2 gan gin VERB bracis-28410 396 3 for for ADP bracis-28410 396 4 robust robust ADJ bracis-28410 396 5 face face NOUN bracis-28410 396 6 super super NOUN bracis-28410 396 7 - - NOUN bracis-28410 396 8 resolution resolution NOUN bracis-28410 396 9 . . PUNCT bracis-28410 397 1 multim multim ADJ bracis-28410 397 2 . . PUNCT bracis-28410 397 3 tools tool NOUN bracis-28410 397 4 appl appl NOUN bracis-28410 397 5 . . PUNCT bracis-28410 398 1 ( ( PUNCT bracis-28410 398 2 2020 2020 NUM bracis-28410 398 3 ) ) PUNCT bracis-28410 398 4 google google PROPN bracis-28410 398 5 scholar scholar NOUN bracis-28410 398 6     SPACE bracis-28410 398 7 martin martin PROPN bracis-28410 398 8 , , PUNCT bracis-28410 398 9 d. d. PROPN bracis-28410 398 10 , , PUNCT bracis-28410 398 11 fowlkes fowlke NOUN bracis-28410 398 12 , , PUNCT bracis-28410 398 13 c. c. NOUN bracis-28410 398 14 , , PUNCT bracis-28410 398 15 tal tal PROPN bracis-28410 398 16 , , PUNCT bracis-28410 398 17 d. d. PROPN bracis-28410 398 18 , , PUNCT bracis-28410 398 19 malik malik PROPN bracis-28410 398 20 , , PUNCT bracis-28410 398 21 j. j. PROPN bracis-28410 398 22 : : PUNCT bracis-28410 398 23 a a DET bracis-28410 398 24 database database NOUN bracis-28410 398 25 of of ADP bracis-28410 398 26 human human ADJ bracis-28410 398 27 segmented segment VERB bracis-28410 398 28 natural natural ADJ bracis-28410 398 29 images image NOUN bracis-28410 398 30 and and CCONJ bracis-28410 398 31 its its PRON bracis-28410 398 32 application application NOUN bracis-28410 398 33 to to ADP bracis-28410 398 34 evaluating evaluate VERB bracis-28410 398 35 segmentation segmentation NOUN bracis-28410 398 36 algorithms algorithm NOUN bracis-28410 398 37 and and CCONJ bracis-28410 398 38 measuring measure VERB bracis-28410 398 39 ecological ecological ADJ bracis-28410 398 40 statistics statistic NOUN bracis-28410 398 41 . . PUNCT bracis-28410 399 1 in in ADP bracis-28410 399 2 : : PUNCT bracis-28410 399 3 iccv iccv NOUN bracis-28410 399 4 , , PUNCT bracis-28410 399 5 vol vol NOUN bracis-28410 399 6 . . PROPN bracis-28410 399 7 2 2 NUM bracis-28410 399 8 , , PUNCT bracis-28410 399 9 pp pp ADJ bracis-28410 399 10 . . PUNCT bracis-28410 400 1 416–423 416–423 NUM bracis-28410 400 2 ( ( PUNCT bracis-28410 400 3 2001 2001 NUM bracis-28410 400 4 ) ) PUNCT bracis-28410 400 5 google google PROPN bracis-28410 400 6 scholar scholar NOUN bracis-28410 400 7     SPACE bracis-28410 400 8 nasrollahi nasrollahi NOUN bracis-28410 400 9 , , PUNCT bracis-28410 400 10 k. k. PROPN bracis-28410 400 11 , , PUNCT bracis-28410 400 12 moeslund moeslund ADV bracis-28410 400 13 , , PUNCT bracis-28410 400 14 t.b t.b PROPN bracis-28410 400 15 . . PUNCT bracis-28410 400 16 : : PUNCT bracis-28410 400 17 super super NOUN bracis-28410 400 18 - - NOUN bracis-28410 400 19 resolution resolution NOUN bracis-28410 400 20 : : PUNCT bracis-28410 400 21 a a DET bracis-28410 400 22 comprehensive comprehensive ADJ bracis-28410 400 23 survey survey NOUN bracis-28410 400 24 . . PUNCT bracis-28410 401 1 mach mach PROPN bracis-28410 401 2 . . PUNCT bracis-28410 402 1 vis vis X bracis-28410 402 2 . . PUNCT bracis-28410 402 3 appl appl PROPN bracis-28410 402 4 . . PUNCT bracis-28410 403 1 25(6 25(6 NUM bracis-28410 403 2 ) ) PUNCT bracis-28410 403 3 , , PUNCT bracis-28410 403 4 1423–1468 1423–1468 NUM bracis-28410 403 5 ( ( PUNCT bracis-28410 403 6 2014 2014 NUM bracis-28410 403 7 ) ) PUNCT bracis-28410 403 8 article article NOUN bracis-28410 403 9     SPACE bracis-28410 403 10 google google PROPN bracis-28410 403 11 scholar scholar NOUN bracis-28410 403 12     SPACE bracis-28410 403 13 odena odena NOUN bracis-28410 403 14 , , PUNCT bracis-28410 403 15 a. a. NOUN bracis-28410 403 16 , , PUNCT bracis-28410 403 17 dumoulin dumoulin PROPN bracis-28410 403 18 , , PUNCT bracis-28410 403 19 v. v. PROPN bracis-28410 403 20 , , PUNCT bracis-28410 403 21 olah olah PROPN bracis-28410 403 22 , , PUNCT bracis-28410 403 23 c. c. NOUN bracis-28410 403 24 : : PUNCT bracis-28410 403 25 deconvolution deconvolution NOUN bracis-28410 403 26 and and CCONJ bracis-28410 403 27 checkerboard checkerboard NOUN bracis-28410 403 28 artifacts artifact NOUN bracis-28410 403 29 . . PUNCT bracis-28410 404 1 distill distill VERB bracis-28410 404 2 ( ( PUNCT bracis-28410 404 3 2016 2016 NUM bracis-28410 404 4 ) ) PUNCT bracis-28410 404 5 . . PUNCT bracis-28410 405 1 http://distill.pub/2016/deconv-checkerboard http://distill.pub/2016/deconv-checkerboard PROPN bracis-28410 405 2 pandey pandey PROPN bracis-28410 405 3 , , PUNCT bracis-28410 405 4 r.k r.k PROPN bracis-28410 405 5 . . PROPN bracis-28410 405 6 , , PUNCT bracis-28410 405 7 saha saha PROPN bracis-28410 405 8 , , PUNCT bracis-28410 405 9 n. n. PROPN bracis-28410 405 10 , , PUNCT bracis-28410 405 11 karmakar karmakar PROPN bracis-28410 405 12 , , PUNCT bracis-28410 405 13 s. s. PROPN bracis-28410 405 14 , , PUNCT bracis-28410 405 15 ramakrishnan ramakrishnan PROPN bracis-28410 405 16 , , PUNCT bracis-28410 405 17 a.g a.g PROPN bracis-28410 405 18 . . PROPN bracis-28410 405 19 : : PUNCT bracis-28410 406 1 msce msce PROPN bracis-28410 406 2 : : PUNCT bracis-28410 406 3 an an DET bracis-28410 406 4 edge edge NOUN bracis-28410 406 5 preserving preserve VERB bracis-28410 406 6 robust robust ADJ bracis-28410 406 7 loss loss NOUN bracis-28410 406 8 function function NOUN bracis-28410 406 9 for for ADP bracis-28410 406 10 improving improve VERB bracis-28410 406 11 super super ADJ bracis-28410 406 12 - - ADJ bracis-28410 406 13 resolution resolution ADJ bracis-28410 406 14 algorithms algorithm NOUN bracis-28410 406 15 . . PUNCT bracis-28410 407 1 arxiv arxiv PROPN bracis-28410 407 2 preprint preprint VERB bracis-28410 407 3 arxiv:1809.00961 arxiv:1809.00961 PROPN bracis-28410 407 4 ( ( PUNCT bracis-28410 407 5 2018 2018 NUM bracis-28410 407 6 ) ) PUNCT bracis-28410 407 7 ren ren PROPN bracis-28410 407 8 , , PUNCT bracis-28410 407 9 h. h. PROPN bracis-28410 407 10 , , PUNCT bracis-28410 407 11 kheradmand kheradmand PROPN bracis-28410 407 12 , , PUNCT bracis-28410 407 13 a. a. PROPN bracis-28410 407 14 , , PUNCT bracis-28410 407 15 el el PROPN bracis-28410 407 16 - - PUNCT bracis-28410 407 17 khamy khamy NOUN bracis-28410 407 18 , , PUNCT bracis-28410 407 19 m. m. NOUN bracis-28410 407 20 , , PUNCT bracis-28410 407 21 wang wang PROPN bracis-28410 407 22 , , PUNCT bracis-28410 407 23 s. s. PROPN bracis-28410 407 24 , , PUNCT bracis-28410 407 25 bai bai PROPN bracis-28410 407 26 , , PUNCT bracis-28410 407 27 d. d. PROPN bracis-28410 407 28 , , PUNCT bracis-28410 407 29 lee lee PROPN bracis-28410 407 30 , , PUNCT bracis-28410 407 31 j. j. PROPN bracis-28410 407 32 : : PUNCT bracis-28410 407 33 real real ADJ bracis-28410 407 34 - - PUNCT bracis-28410 407 35 world world NOUN bracis-28410 407 36 super super NOUN bracis-28410 407 37 - - NOUN bracis-28410 407 38 resolution resolution NOUN bracis-28410 407 39 using use VERB bracis-28410 407 40 generative generative ADJ bracis-28410 407 41 adversarial adversarial ADJ bracis-28410 407 42 networks network NOUN bracis-28410 407 43 . . PUNCT bracis-28410 408 1 in in ADP bracis-28410 408 2 : : PUNCT bracis-28410 408 3 cvpr cvpr NOUN bracis-28410 408 4 workshops workshop NOUN bracis-28410 408 5 , , PUNCT bracis-28410 408 6 pp pp ADV bracis-28410 408 7 . . PUNCT bracis-28410 409 1 1–8 1–8 NUM bracis-28410 409 2 ( ( PUNCT bracis-28410 409 3 2020 2020 NUM bracis-28410 409 4 ) ) PUNCT bracis-28410 409 5 google google PROPN bracis-28410 409 6 scholar scholar NOUN bracis-28410 409 7     SPACE bracis-28410 409 8 sabour sabour NOUN bracis-28410 409 9 , , PUNCT bracis-28410 409 10 s. s. PROPN bracis-28410 409 11 , , PUNCT bracis-28410 409 12 frosst frosst ADJ bracis-28410 409 13 , , PUNCT bracis-28410 409 14 n. n. PROPN bracis-28410 409 15 , , PUNCT bracis-28410 409 16 hinton hinton PROPN bracis-28410 409 17 , , PUNCT bracis-28410 409 18 g.e g.e PROPN bracis-28410 409 19 . . PUNCT bracis-28410 409 20 : : PUNCT bracis-28410 410 1 dynamic dynamic ADJ bracis-28410 410 2 routing routing NOUN bracis-28410 410 3 between between ADP bracis-28410 410 4 capsules capsule NOUN bracis-28410 410 5 . . PUNCT bracis-28410 411 1 in in ADP bracis-28410 411 2 : : PUNCT bracis-28410 411 3 neurips neurip NOUN bracis-28410 411 4 , , PUNCT bracis-28410 411 5 pp pp ADV bracis-28410 411 6 . . PUNCT bracis-28410 412 1 3856–3866 3856–3866 NUM bracis-28410 412 2 ( ( PUNCT bracis-28410 412 3 2017 2017 NUM bracis-28410 412 4 ) ) PUNCT bracis-28410 412 5 google google PROPN bracis-28410 412 6 scholar scholar NOUN bracis-28410 412 7     SPACE bracis-28410 412 8 sabour sabour NOUN bracis-28410 412 9 , , PUNCT bracis-28410 412 10 s. s. PROPN bracis-28410 412 11 , , PUNCT bracis-28410 412 12 tagliasacchi tagliasacchi PROPN bracis-28410 412 13 , , PUNCT bracis-28410 412 14 a. a. PROPN bracis-28410 412 15 , , PUNCT bracis-28410 412 16 yazdani yazdani PROPN bracis-28410 412 17 , , PUNCT bracis-28410 412 18 s. s. PROPN bracis-28410 412 19 , , PUNCT bracis-28410 412 20 hinton hinton PROPN bracis-28410 412 21 , , PUNCT bracis-28410 412 22 g.e g.e PROPN bracis-28410 412 23 . . PROPN bracis-28410 412 24 , , PUNCT bracis-28410 412 25 fleet fleet PROPN bracis-28410 412 26 , , PUNCT bracis-28410 412 27 d.j d.j PROPN bracis-28410 412 28 . . PROPN bracis-28410 412 29 : : PUNCT bracis-28410 413 1 unsupervised unsupervised ADJ bracis-28410 413 2 part part NOUN bracis-28410 413 3 representation representation NOUN bracis-28410 413 4 by by ADP bracis-28410 413 5 flow flow NOUN bracis-28410 413 6 capsules capsule NOUN bracis-28410 413 7 . . PUNCT bracis-28410 414 1 in in ADP bracis-28410 414 2 : : PUNCT bracis-28410 414 3 meila meila NOUN bracis-28410 414 4 , , PUNCT bracis-28410 414 5 m. m. NOUN bracis-28410 414 6 , , PUNCT bracis-28410 414 7 zhang zhang PROPN bracis-28410 414 8 , , PUNCT bracis-28410 414 9 t. t. PROPN bracis-28410 414 10 ( ( PUNCT bracis-28410 414 11 eds eds PROPN bracis-28410 414 12 . . PUNCT bracis-28410 414 13 ) ) PUNCT bracis-28410 414 14 icml icml NOUN bracis-28410 414 15 , , PUNCT bracis-28410 414 16 vol vol NOUN bracis-28410 414 17 . . PROPN bracis-28410 414 18 139 139 NUM bracis-28410 414 19 , , PUNCT bracis-28410 414 20 pp pp ADJ bracis-28410 414 21 . . PUNCT bracis-28410 415 1 9213–9223 9213–9223 NUM bracis-28410 415 2 ( ( PUNCT bracis-28410 415 3 2021 2021 NUM bracis-28410 415 4 ) ) PUNCT bracis-28410 415 5 google google NOUN bracis-28410 415 6 scholar scholar NOUN bracis-28410 415 7     SPACE bracis-28410 415 8 salimans saliman NOUN bracis-28410 415 9 , , PUNCT bracis-28410 415 10 t. t. PROPN bracis-28410 415 11 , , PUNCT bracis-28410 415 12 kingma kingma PROPN bracis-28410 415 13 , , PUNCT bracis-28410 415 14 d.p d.p PROPN bracis-28410 415 15 . . PROPN bracis-28410 415 16 : : PUNCT bracis-28410 416 1 weight weight NOUN bracis-28410 416 2 normalization normalization NOUN bracis-28410 416 3 : : PUNCT bracis-28410 416 4 a a DET bracis-28410 416 5 simple simple ADJ bracis-28410 416 6 reparameterization reparameterization NOUN bracis-28410 416 7 to to PART bracis-28410 416 8 accelerate accelerate VERB bracis-28410 416 9 training training NOUN bracis-28410 416 10 of of ADP bracis-28410 416 11 deep deep ADJ bracis-28410 416 12 neural neural ADJ bracis-28410 416 13 networks network NOUN bracis-28410 416 14 . . PUNCT bracis-28410 417 1 in in ADP bracis-28410 417 2 : : PUNCT bracis-28410 417 3 neurips neurip NOUN bracis-28410 417 4 , , PUNCT bracis-28410 417 5 pp pp ADV bracis-28410 417 6 . . PUNCT bracis-28410 418 1 901–909 901–909 X bracis-28410 418 2 . . PUNCT bracis-28410 418 3 curran curran PROPN bracis-28410 418 4 associates associates PROPN bracis-28410 418 5 , , PUNCT bracis-28410 418 6 inc inc PROPN bracis-28410 418 7 . . PROPN bracis-28410 418 8 ( ( PUNCT bracis-28410 418 9 2016 2016 NUM bracis-28410 418 10 ) ) PUNCT bracis-28410 418 11 google google PROPN bracis-28410 418 12 scholar scholar NOUN bracis-28410 418 13     SPACE bracis-28410 418 14 shi shi PROPN bracis-28410 418 15 , , PUNCT bracis-28410 418 16 w. w. PROPN bracis-28410 418 17 , , PUNCT bracis-28410 418 18 et et PROPN bracis-28410 418 19 al al PROPN bracis-28410 418 20 . . PROPN bracis-28410 418 21 : : PUNCT bracis-28410 419 1 real real ADJ bracis-28410 419 2 - - PUNCT bracis-28410 419 3 time time NOUN bracis-28410 419 4 single single ADJ bracis-28410 419 5 image image NOUN bracis-28410 419 6 and and CCONJ bracis-28410 419 7 video video NOUN bracis-28410 419 8 super super NOUN bracis-28410 419 9 - - NOUN bracis-28410 419 10 resolution resolution NOUN bracis-28410 419 11 using use VERB bracis-28410 419 12 an an DET bracis-28410 419 13 efficient efficient ADJ bracis-28410 419 14 sub sub ADJ bracis-28410 419 15 - - ADJ bracis-28410 419 16 pixel pixel ADJ bracis-28410 419 17 convolutional convolutional ADJ bracis-28410 419 18 neural neural ADJ bracis-28410 419 19 network network NOUN bracis-28410 419 20 . . PUNCT bracis-28410 420 1 in in ADP bracis-28410 420 2 : : PUNCT bracis-28410 420 3 cvpr cvpr NOUN bracis-28410 420 4 , , PUNCT bracis-28410 420 5 pp pp ADV bracis-28410 420 6 . . PUNCT bracis-28410 421 1 1–8 1–8 NUM bracis-28410 421 2 ( ( PUNCT bracis-28410 421 3 2016 2016 NUM bracis-28410 421 4 ) ) PUNCT bracis-28410 421 5 google google PROPN bracis-28410 421 6 scholar scholar NOUN bracis-28410 421 7     SPACE bracis-28410 421 8 simonyan simonyan ADJ bracis-28410 421 9 , , PUNCT bracis-28410 421 10 k. k. PROPN bracis-28410 421 11 , , PUNCT bracis-28410 421 12 zisserman zisserman PROPN bracis-28410 421 13 , , PUNCT bracis-28410 421 14 a. a. NOUN bracis-28410 421 15 : : PUNCT bracis-28410 421 16 very very ADV bracis-28410 421 17 deep deep ADJ bracis-28410 421 18 convolutional convolutional ADJ bracis-28410 421 19 networks network NOUN bracis-28410 421 20 for for ADP bracis-28410 421 21 large large ADJ bracis-28410 421 22 - - PUNCT bracis-28410 421 23 scale scale NOUN bracis-28410 421 24 image image NOUN bracis-28410 421 25 recognition recognition NOUN bracis-28410 421 26 . . PUNCT bracis-28410 422 1 in in ADP bracis-28410 422 2 : : PUNCT bracis-28410 422 3 iclr iclr NOUN bracis-28410 422 4 ( ( PUNCT bracis-28410 422 5 2015 2015 NUM bracis-28410 422 6 ) ) PUNCT bracis-28410 422 7 google google PROPN bracis-28410 422 8 scholar scholar NOUN bracis-28410 422 9     SPACE bracis-28410 422 10 singh singh PROPN bracis-28410 422 11 , , PUNCT bracis-28410 422 12 m. m. NOUN bracis-28410 422 13 , , PUNCT bracis-28410 422 14 nagpal nagpal NOUN bracis-28410 422 15 , , PUNCT bracis-28410 422 16 s. s. PROPN bracis-28410 422 17 , , PUNCT bracis-28410 422 18 singh singh PROPN bracis-28410 422 19 , , PUNCT bracis-28410 422 20 r. r. PROPN bracis-28410 422 21 , , PUNCT bracis-28410 422 22 vatsa vatsa NOUN bracis-28410 422 23 , , PUNCT bracis-28410 422 24 m. m. NOUN bracis-28410 422 25 : : PUNCT bracis-28410 422 26 dual dual ADJ bracis-28410 422 27 directed direct VERB bracis-28410 422 28 capsule capsule NOUN bracis-28410 422 29 network network NOUN bracis-28410 422 30 for for ADP bracis-28410 422 31 very very ADV bracis-28410 422 32 low low ADJ bracis-28410 422 33 resolution resolution NOUN bracis-28410 422 34 image image NOUN bracis-28410 422 35 recognition recognition NOUN bracis-28410 422 36 . . PUNCT bracis-28410 423 1 in in ADP bracis-28410 423 2 : : PUNCT bracis-28410 423 3 iccv iccv PROPN bracis-28410 423 4 , , PUNCT bracis-28410 423 5 pp pp ADV bracis-28410 423 6 . . PUNCT bracis-28410 424 1 1–8 1–8 NUM bracis-28410 424 2 ( ( PUNCT bracis-28410 424 3 2019 2019 NUM bracis-28410 424 4 ) ) PUNCT bracis-28410 424 5 google google PROPN bracis-28410 424 6 scholar scholar NOUN bracis-28410 424 7     SPACE bracis-28410 424 8 sobel sobel NOUN bracis-28410 424 9 , , PUNCT bracis-28410 424 10 i. i. PROPN bracis-28410 424 11 , , PUNCT bracis-28410 424 12 feldman feldman PROPN bracis-28410 424 13 , , PUNCT bracis-28410 424 14 g. g. PROPN bracis-28410 424 15 : : PUNCT bracis-28410 424 16 a a DET bracis-28410 424 17 \(3\times \(3\time NOUN bracis-28410 424 18 3\ 3\ NUM bracis-28410 424 19 ) ) PUNCT bracis-28410 424 20 isotropic isotropic ADJ bracis-28410 424 21 gradient gradient NOUN bracis-28410 424 22 operator operator NOUN bracis-28410 424 23 for for ADP bracis-28410 424 24 image image NOUN bracis-28410 424 25 processing processing NOUN bracis-28410 424 26 ( ( PUNCT bracis-28410 424 27 1968 1968 NUM bracis-28410 424 28 ) ) PUNCT bracis-28410 424 29 . . PUNCT bracis-28410 425 1 talk talk VERB bracis-28410 425 2 at at ADP bracis-28410 425 3 the the DET bracis-28410 425 4 stanford stanford PROPN bracis-28410 425 5 artificial artificial PROPN bracis-28410 425 6 project project NOUN bracis-28410 425 7 google google PROPN bracis-28410 425 8 scholar scholar NOUN bracis-28410 425 9     SPACE bracis-28410 425 10 szegedy szegedy PROPN bracis-28410 425 11 , , PUNCT bracis-28410 425 12 c. c. NOUN bracis-28410 425 13 , , PUNCT bracis-28410 425 14 ioffe ioffe PROPN bracis-28410 425 15 , , PUNCT bracis-28410 425 16 s. s. PROPN bracis-28410 425 17 , , PUNCT bracis-28410 425 18 vanhoucke vanhoucke PROPN bracis-28410 425 19 , , PUNCT bracis-28410 425 20 v. v. ADV bracis-28410 425 21 , , PUNCT bracis-28410 425 22 alemi alemi PROPN bracis-28410 425 23 , , PUNCT bracis-28410 425 24 a.a a.a PROPN bracis-28410 425 25 . . PROPN bracis-28410 425 26 : : PUNCT bracis-28410 425 27 inception inception NOUN bracis-28410 425 28 - - PUNCT bracis-28410 425 29 v4 v4 NOUN bracis-28410 425 30 , , PUNCT bracis-28410 425 31 inception inception NOUN bracis-28410 425 32 - - PUNCT bracis-28410 425 33 resnet resnet NOUN bracis-28410 425 34 and and CCONJ bracis-28410 425 35 the the DET bracis-28410 425 36 impact impact NOUN bracis-28410 425 37 of of ADP bracis-28410 425 38 residual residual ADJ bracis-28410 425 39 connections connection NOUN bracis-28410 425 40 on on ADP bracis-28410 425 41 learning learn VERB bracis-28410 425 42 . . PUNCT bracis-28410 426 1 in in ADP bracis-28410 426 2 : : PUNCT bracis-28410 426 3 aaai aaai PROPN bracis-28410 426 4 , , PUNCT bracis-28410 426 5 pp pp ADJ bracis-28410 426 6 . . PUNCT bracis-28410 427 1 4278–4284 4278–4284 NUM bracis-28410 427 2 ( ( PUNCT bracis-28410 427 3 2017 2017 NUM bracis-28410 427 4 ) ) PUNCT bracis-28410 427 5 google google PROPN bracis-28410 427 6 scholar scholar NOUN bracis-28410 427 7     SPACE bracis-28410 427 8 szegedy szegedy PROPN bracis-28410 427 9 , , PUNCT bracis-28410 427 10 c. c. PROPN bracis-28410 427 11 , , PUNCT bracis-28410 427 12 et et PROPN bracis-28410 427 13 al al PROPN bracis-28410 427 14 . . PROPN bracis-28410 427 15 : : PUNCT bracis-28410 427 16 going go VERB bracis-28410 427 17 deeper deeply ADV bracis-28410 427 18 with with ADP bracis-28410 427 19 convolutions convolution NOUN bracis-28410 427 20 . . PUNCT bracis-28410 428 1 in in ADP bracis-28410 428 2 : : PUNCT bracis-28410 428 3 cvpr cvpr NOUN bracis-28410 428 4 , , PUNCT bracis-28410 428 5 pp pp ADV bracis-28410 428 6 . . PUNCT bracis-28410 429 1 1–8 1–8 NUM bracis-28410 429 2 ( ( PUNCT bracis-28410 429 3 2015 2015 NUM bracis-28410 429 4 ) ) PUNCT bracis-28410 429 5 google google PROPN bracis-28410 429 6 scholar scholar NOUN bracis-28410 429 7     SPACE bracis-28410 429 8 timofte timofte NOUN bracis-28410 429 9 , , PUNCT bracis-28410 429 10 r. r. PROPN bracis-28410 429 11 , , PUNCT bracis-28410 429 12 et et PROPN bracis-28410 429 13 al al PROPN bracis-28410 429 14 . . PUNCT bracis-28410 429 15 : : PUNCT bracis-28410 429 16 ntire ntire NOUN bracis-28410 429 17 2017 2017 NUM bracis-28410 429 18 challenge challenge NOUN bracis-28410 429 19 on on ADP bracis-28410 429 20 single single ADJ bracis-28410 429 21 image image NOUN bracis-28410 429 22 super super NOUN bracis-28410 429 23 - - NOUN bracis-28410 429 24 resolution resolution NOUN bracis-28410 429 25 : : PUNCT bracis-28410 429 26 methods method NOUN bracis-28410 429 27 and and CCONJ bracis-28410 429 28 results result NOUN bracis-28410 429 29 . . PUNCT bracis-28410 430 1 in in ADP bracis-28410 430 2 : : PUNCT bracis-28410 430 3 cvpr cvpr NOUN bracis-28410 430 4 workshops workshop NOUN bracis-28410 430 5 , , PUNCT bracis-28410 430 6 pp pp ADV bracis-28410 430 7 . . PUNCT bracis-28410 431 1 1110–1121 1110–1121 NUM bracis-28410 431 2 ( ( PUNCT bracis-28410 431 3 2017 2017 NUM bracis-28410 431 4 ) ) PUNCT bracis-28410 431 5 google google PROPN bracis-28410 431 6 scholar scholar NOUN bracis-28410 431 7     SPACE bracis-28410 431 8 timofte timofte NOUN bracis-28410 431 9 , , PUNCT bracis-28410 431 10 r. r. PROPN bracis-28410 431 11 , , PUNCT bracis-28410 431 12 gu gu PROPN bracis-28410 431 13 , , PUNCT bracis-28410 431 14 s. s. PROPN bracis-28410 431 15 , , PUNCT bracis-28410 431 16 wu wu PROPN bracis-28410 431 17 , , PUNCT bracis-28410 431 18 j. j. PROPN bracis-28410 431 19 , , PUNCT bracis-28410 431 20 van van PROPN bracis-28410 431 21 gool gool PROPN bracis-28410 431 22 , , PUNCT bracis-28410 431 23 l. l. PROPN bracis-28410 431 24 : : PUNCT bracis-28410 431 25 ntire ntire NOUN bracis-28410 431 26 2018 2018 NUM bracis-28410 431 27 challenge challenge NOUN bracis-28410 431 28 on on ADP bracis-28410 431 29 single single ADJ bracis-28410 431 30 image image NOUN bracis-28410 431 31 super super NOUN bracis-28410 431 32 - - NOUN bracis-28410 431 33 resolution resolution NOUN bracis-28410 431 34 : : PUNCT bracis-28410 431 35 methods method NOUN bracis-28410 431 36 and and CCONJ bracis-28410 431 37 results result NOUN bracis-28410 431 38 . . PUNCT bracis-28410 432 1 in in ADP bracis-28410 432 2 : : PUNCT bracis-28410 432 3 cvpr cvpr NOUN bracis-28410 432 4 workshops workshop NOUN bracis-28410 432 5 , , PUNCT bracis-28410 432 6 pp pp ADV bracis-28410 432 7 . . PUNCT bracis-28410 433 1 1–17 1–17 PROPN bracis-28410 433 2 ( ( PUNCT bracis-28410 433 3 2018 2018 NUM bracis-28410 433 4 ) ) PUNCT bracis-28410 433 5 google google PROPN bracis-28410 433 6 scholar scholar NOUN bracis-28410 433 7     SPACE bracis-28410 433 8 veit veit PROPN bracis-28410 433 9 , , PUNCT bracis-28410 433 10 a. a. PROPN bracis-28410 433 11 , , PUNCT bracis-28410 433 12 wilber wilber PROPN bracis-28410 433 13 , , PUNCT bracis-28410 433 14 m.j m.j PROPN bracis-28410 433 15 . . PROPN bracis-28410 433 16 , , PUNCT bracis-28410 433 17 belongie belongie PROPN bracis-28410 433 18 , , PUNCT bracis-28410 433 19 s. s. PROPN bracis-28410 433 20 : : PUNCT bracis-28410 433 21 residual residual ADJ bracis-28410 433 22 networks network NOUN bracis-28410 433 23 behave behave VERB bracis-28410 433 24 like like ADP bracis-28410 433 25 ensembles ensemble NOUN bracis-28410 433 26 of of ADP bracis-28410 433 27 relatively relatively ADV bracis-28410 433 28 shallow shallow ADJ bracis-28410 433 29 networks network NOUN bracis-28410 433 30 . . PUNCT bracis-28410 434 1 in in ADP bracis-28410 434 2 : : PUNCT bracis-28410 434 3 neurips neurip NOUN bracis-28410 434 4 , , PUNCT bracis-28410 434 5 pp pp ADV bracis-28410 434 6 . . PUNCT bracis-28410 435 1 550–558 550–558 NUM bracis-28410 435 2 ( ( PUNCT bracis-28410 435 3 2016 2016 NUM bracis-28410 435 4 ) ) PUNCT bracis-28410 435 5 google google PROPN bracis-28410 435 6 scholar scholar NOUN bracis-28410 435 7     SPACE bracis-28410 435 8 wang wang PROPN bracis-28410 435 9 , , PUNCT bracis-28410 435 10 x. x. PROPN bracis-28410 435 11 , , PUNCT bracis-28410 435 12 et et PROPN bracis-28410 435 13 al al PROPN bracis-28410 435 14 . . PROPN bracis-28410 435 15 : : PUNCT bracis-28410 435 16 esrgan esrgan PROPN bracis-28410 435 17 : : PUNCT bracis-28410 435 18 enhanced enhance VERB bracis-28410 435 19 super super ADJ bracis-28410 435 20 - - ADJ bracis-28410 435 21 resolution resolution ADJ bracis-28410 435 22 generative generative ADJ bracis-28410 435 23 adversarial adversarial ADJ bracis-28410 435 24 networks network NOUN bracis-28410 435 25 . . PUNCT bracis-28410 436 1 in in ADP bracis-28410 436 2 : : PUNCT bracis-28410 436 3 eccv eccv ADJ bracis-28410 436 4 , , PUNCT bracis-28410 436 5 pp pp ADJ bracis-28410 436 6 . . PUNCT bracis-28410 437 1 63–79 63–79 NUM bracis-28410 437 2 ( ( PUNCT bracis-28410 437 3 2019 2019 NUM bracis-28410 437 4 ) ) PUNCT bracis-28410 437 5 google google PROPN bracis-28410 437 6 scholar scholar NOUN bracis-28410 437 7     SPACE bracis-28410 437 8 wang wang PROPN bracis-28410 437 9 , , PUNCT bracis-28410 437 10 z. z. PROPN bracis-28410 437 11 , , PUNCT bracis-28410 437 12 liu liu PROPN bracis-28410 437 13 , , PUNCT bracis-28410 437 14 d. d. PROPN bracis-28410 437 15 , , PUNCT bracis-28410 437 16 yang yang PROPN bracis-28410 437 17 , , PUNCT bracis-28410 437 18 j. j. PROPN bracis-28410 437 19 , , PUNCT bracis-28410 437 20 han han PROPN bracis-28410 437 21 , , PUNCT bracis-28410 437 22 w. w. PROPN bracis-28410 437 23 , , PUNCT bracis-28410 437 24 huang huang PROPN bracis-28410 437 25 , , PUNCT bracis-28410 437 26 t. t. PROPN bracis-28410 437 27 : : PUNCT bracis-28410 437 28 deep deep ADJ bracis-28410 437 29 networks network NOUN bracis-28410 437 30 for for ADP bracis-28410 437 31 image image NOUN bracis-28410 437 32 super super NOUN bracis-28410 437 33 - - NOUN bracis-28410 437 34 resolution resolution NOUN bracis-28410 437 35 with with ADP bracis-28410 437 36 sparse sparse ADJ bracis-28410 437 37 prior prior ADV bracis-28410 437 38 . . PUNCT bracis-28410 438 1 in in ADP bracis-28410 438 2 : : PUNCT bracis-28410 438 3 iccv iccv PROPN bracis-28410 438 4 , , PUNCT bracis-28410 438 5 pp pp ADJ bracis-28410 438 6 . . PUNCT bracis-28410 439 1 370–378 370–378 NUM bracis-28410 439 2 ( ( PUNCT bracis-28410 439 3 2015 2015 NUM bracis-28410 439 4 ) ) PUNCT bracis-28410 439 5 google google PROPN bracis-28410 439 6 scholar scholar NOUN bracis-28410 439 7     SPACE bracis-28410 439 8 wang wang PROPN bracis-28410 439 9 , , PUNCT bracis-28410 439 10 z. z. PROPN bracis-28410 439 11 , , PUNCT bracis-28410 439 12 simoncelli simoncelli PROPN bracis-28410 439 13 , , PUNCT bracis-28410 439 14 e.p e.p PROPN bracis-28410 439 15 . . PROPN bracis-28410 439 16 , , PUNCT bracis-28410 439 17 bovik bovik NOUN bracis-28410 439 18 , , PUNCT bracis-28410 439 19 a.c a.c PROPN bracis-28410 439 20 . . PROPN bracis-28410 439 21 : : PUNCT bracis-28410 440 1 multiscale multiscale ADJ bracis-28410 440 2 structural structural ADJ bracis-28410 440 3 similarity similarity NOUN bracis-28410 440 4 for for ADP bracis-28410 440 5 image image NOUN bracis-28410 440 6 quality quality NOUN bracis-28410 440 7 assessment assessment NOUN bracis-28410 440 8 . . PUNCT bracis-28410 441 1 in in ADP bracis-28410 441 2 : : PUNCT bracis-28410 441 3 the the DET bracis-28410 441 4 thirty thirty NUM bracis-28410 441 5 - - PUNCT bracis-28410 441 6 seventh seventh NOUN bracis-28410 441 7 asilomar asilomar NOUN bracis-28410 441 8 conference conference NOUN bracis-28410 441 9 on on ADP bracis-28410 441 10 signals signal NOUN bracis-28410 441 11 , , PUNCT bracis-28410 441 12 systems system NOUN bracis-28410 441 13 computers computer NOUN bracis-28410 441 14 , , PUNCT bracis-28410 441 15 vol vol NOUN bracis-28410 441 16 . . PROPN bracis-28410 441 17 2 2 NUM bracis-28410 441 18 , , PUNCT bracis-28410 441 19 pp pp ADJ bracis-28410 441 20 . . PUNCT bracis-28410 442 1 1398–1402 1398–1402 NUM bracis-28410 442 2 ( ( PUNCT bracis-28410 442 3 2003 2003 NUM bracis-28410 442 4 ) ) PUNCT bracis-28410 442 5 google google PROPN bracis-28410 442 6 scholar scholar NOUN bracis-28410 442 7     SPACE bracis-28410 442 8 wang wang PROPN bracis-28410 442 9 , , PUNCT bracis-28410 442 10 z. z. PROPN bracis-28410 442 11 , , PUNCT bracis-28410 442 12 chen chen PROPN bracis-28410 442 13 , , PUNCT bracis-28410 442 14 j. j. PROPN bracis-28410 442 15 , , PUNCT bracis-28410 442 16 hoi hoi PROPN bracis-28410 442 17 , , PUNCT bracis-28410 442 18 s.c.h s.c.h NOUN bracis-28410 442 19 . . PUNCT bracis-28410 442 20 : : PUNCT bracis-28410 443 1 deep deep ADJ bracis-28410 443 2 learning learn VERB bracis-28410 443 3 for for ADP bracis-28410 443 4 image image NOUN bracis-28410 443 5 super super NOUN bracis-28410 443 6 - - NOUN bracis-28410 443 7 resolution resolution NOUN bracis-28410 443 8 : : PUNCT bracis-28410 443 9 a a DET bracis-28410 443 10 survey survey NOUN bracis-28410 443 11 . . PUNCT bracis-28410 444 1 trans tran NOUN bracis-28410 444 2 . . PUNCT bracis-28410 444 3 pattern pattern PROPN bracis-28410 444 4 anal anal PROPN bracis-28410 444 5 . . PUNCT bracis-28410 445 1 mach mach PROPN bracis-28410 445 2 . . PUNCT bracis-28410 446 1 intell intell PROPN bracis-28410 446 2 . . PUNCT bracis-28410 447 1 43(10 43(10 NUM bracis-28410 447 2 ) ) PUNCT bracis-28410 447 3 , , PUNCT bracis-28410 447 4 3365–3387 3365–3387 NUM bracis-28410 447 5 ( ( PUNCT bracis-28410 447 6 2021 2021 NUM bracis-28410 447 7 ) ) PUNCT bracis-28410 447 8 article article NOUN bracis-28410 447 9     SPACE bracis-28410 447 10 google google PROPN bracis-28410 447 11 scholar scholar NOUN bracis-28410 447 12     SPACE bracis-28410 447 13 wang wang PROPN bracis-28410 447 14 , , PUNCT bracis-28410 447 15 z. z. PROPN bracis-28410 447 16 , , PUNCT bracis-28410 447 17 bovik bovik NOUN bracis-28410 447 18 , , PUNCT bracis-28410 447 19 a.c a.c PROPN bracis-28410 447 20 . . PROPN bracis-28410 447 21 , , PUNCT bracis-28410 447 22 sheikh sheikh PROPN bracis-28410 447 23 , , PUNCT bracis-28410 447 24 h.r h.r PROPN bracis-28410 447 25 . . PROPN bracis-28410 447 26 , , PUNCT bracis-28410 447 27 simoncelli simoncelli PROPN bracis-28410 447 28 , , PUNCT bracis-28410 447 29 e.p e.p PROPN bracis-28410 447 30 . . PROPN bracis-28410 447 31 : : PUNCT bracis-28410 447 32 image image NOUN bracis-28410 447 33 quality quality NOUN bracis-28410 447 34 assessment assessment NOUN bracis-28410 447 35 : : PUNCT bracis-28410 447 36 from from ADP bracis-28410 447 37 error error NOUN bracis-28410 447 38 visibility visibility NOUN bracis-28410 447 39 to to ADP bracis-28410 447 40 structural structural ADJ bracis-28410 447 41 similarity similarity NOUN bracis-28410 447 42 . . PUNCT bracis-28410 448 1 trans trans PROPN bracis-28410 448 2 . . PUNCT bracis-28410 448 3 image image NOUN bracis-28410 448 4 process process NOUN bracis-28410 448 5 . . PUNCT bracis-28410 449 1 13(4 13(4 NUM bracis-28410 449 2 ) ) PUNCT bracis-28410 449 3 , , PUNCT bracis-28410 449 4 600–612 600–612 NUM bracis-28410 449 5 ( ( PUNCT bracis-28410 449 6 2004 2004 NUM bracis-28410 449 7 ) ) PUNCT bracis-28410 449 8 article article NOUN bracis-28410 449 9     SPACE bracis-28410 449 10 google google PROPN bracis-28410 449 11 scholar scholar NOUN bracis-28410 449 12     SPACE bracis-28410 449 13 xu xu PROPN bracis-28410 449 14 , , PUNCT bracis-28410 449 15 j. j. PROPN bracis-28410 449 16 , , PUNCT bracis-28410 449 17 zhao zhao PROPN bracis-28410 449 18 , , PUNCT bracis-28410 449 19 y. y. PROPN bracis-28410 449 20 , , PUNCT bracis-28410 449 21 dong dong PROPN bracis-28410 449 22 , , PUNCT bracis-28410 449 23 y. y. PROPN bracis-28410 449 24 , , PUNCT bracis-28410 449 25 bai bai PROPN bracis-28410 449 26 , , PUNCT bracis-28410 449 27 h. h. PROPN bracis-28410 449 28 : : PUNCT bracis-28410 449 29 fast fast ADJ bracis-28410 449 30 and and CCONJ bracis-28410 449 31 accurate accurate ADJ bracis-28410 449 32 image image NOUN bracis-28410 449 33 super super NOUN bracis-28410 449 34 - - NOUN bracis-28410 449 35 resolution resolution NOUN bracis-28410 449 36 using use VERB bracis-28410 449 37 a a DET bracis-28410 449 38 combined combined ADJ bracis-28410 449 39 loss loss NOUN bracis-28410 449 40 . . PUNCT bracis-28410 450 1 in in ADP bracis-28410 450 2 : : PUNCT bracis-28410 450 3 cvpr cvpr NOUN bracis-28410 450 4 workshops workshop NOUN bracis-28410 450 5 , , PUNCT bracis-28410 450 6 pp pp ADV bracis-28410 450 7 . . PUNCT bracis-28410 451 1 1093–1099 1093–1099 NUM bracis-28410 451 2 ( ( PUNCT bracis-28410 451 3 2017 2017 NUM bracis-28410 451 4 ) ) PUNCT bracis-28410 451 5 google google PROPN bracis-28410 451 6 scholar scholar NOUN bracis-28410 451 7     SPACE bracis-28410 451 8 yang yang PROPN bracis-28410 451 9 , , PUNCT bracis-28410 451 10 j. j. PROPN bracis-28410 451 11 , , PUNCT bracis-28410 451 12 wright wright PROPN bracis-28410 451 13 , , PUNCT bracis-28410 451 14 j. j. PROPN bracis-28410 451 15 , , PUNCT bracis-28410 451 16 huang huang PROPN bracis-28410 451 17 , , PUNCT bracis-28410 451 18 t.s t.s PROPN bracis-28410 451 19 . . PROPN bracis-28410 451 20 , , PUNCT bracis-28410 451 21 ma ma PROPN bracis-28410 451 22 , , PUNCT bracis-28410 451 23 y. y. NOUN bracis-28410 451 24 : : PUNCT bracis-28410 451 25 image image NOUN bracis-28410 451 26 super super NOUN bracis-28410 451 27 - - NOUN bracis-28410 451 28 resolution resolution NOUN bracis-28410 451 29 via via ADP bracis-28410 451 30 sparse sparse ADJ bracis-28410 451 31 representation representation NOUN bracis-28410 451 32 . . PUNCT bracis-28410 452 1 trans trans PROPN bracis-28410 452 2 . . PUNCT bracis-28410 452 3 image image NOUN bracis-28410 452 4 process process NOUN bracis-28410 452 5 . . PUNCT bracis-28410 453 1 19(11 19(11 NUM bracis-28410 453 2 ) ) PUNCT bracis-28410 453 3 , , PUNCT bracis-28410 453 4 2861–2873 2861–2873 NUM bracis-28410 453 5 ( ( PUNCT bracis-28410 453 6 2010 2010 NUM bracis-28410 453 7 ) ) PUNCT bracis-28410 453 8 article article NOUN bracis-28410 453 9     SPACE bracis-28410 453 10 mathscinet mathscinet NOUN bracis-28410 453 11     SPACE bracis-28410 453 12 math math NOUN bracis-28410 453 13     SPACE bracis-28410 453 14 google google PROPN bracis-28410 453 15 scholar scholar NOUN bracis-28410 453 16     SPACE bracis-28410 453 17 yu yu PROPN bracis-28410 453 18 , , PUNCT bracis-28410 453 19 c. c. PROPN bracis-28410 453 20 , , PUNCT bracis-28410 453 21 zhu zhu PROPN bracis-28410 453 22 , , PUNCT bracis-28410 453 23 x. x. PROPN bracis-28410 453 24 , , PUNCT bracis-28410 453 25 zhang zhang PROPN bracis-28410 453 26 , , PUNCT bracis-28410 453 27 x. x. PROPN bracis-28410 453 28 , , PUNCT bracis-28410 453 29 wang wang PROPN bracis-28410 453 30 , , PUNCT bracis-28410 453 31 z. z. PROPN bracis-28410 453 32 , , PUNCT bracis-28410 453 33 zhang zhang PROPN bracis-28410 453 34 , , PUNCT bracis-28410 453 35 z. z. PROPN bracis-28410 453 36 , , PUNCT bracis-28410 453 37 lei lei PROPN bracis-28410 453 38 , , PUNCT bracis-28410 453 39 z. z. PROPN bracis-28410 453 40 : : PUNCT bracis-28410 453 41 hp hp PROPN bracis-28410 453 42 - - PUNCT bracis-28410 453 43 capsule capsule NOUN bracis-28410 453 44 : : PUNCT bracis-28410 453 45 unsupervised unsupervised ADJ bracis-28410 453 46 face face NOUN bracis-28410 453 47 part part NOUN bracis-28410 453 48 discovery discovery NOUN bracis-28410 453 49 by by ADP bracis-28410 453 50 hierarchical hierarchical ADJ bracis-28410 453 51 parsing parse VERB bracis-28410 453 52 capsule capsule NOUN bracis-28410 453 53 network network NOUN bracis-28410 453 54 . . PUNCT bracis-28410 454 1 in in ADP bracis-28410 454 2 : : PUNCT bracis-28410 454 3 cvpr cvpr NOUN bracis-28410 454 4 , , PUNCT bracis-28410 454 5 pp pp ADV bracis-28410 454 6 . . PUNCT bracis-28410 455 1 4022–4031 4022–4031 NOUN bracis-28410 455 2 ( ( PUNCT bracis-28410 455 3 2022 2022 NUM bracis-28410 455 4 ) ) PUNCT bracis-28410 455 5 google google PROPN bracis-28410 455 6 scholar scholar NOUN bracis-28410 455 7     SPACE bracis-28410 455 8 yu yu PROPN bracis-28410 455 9 , , PUNCT bracis-28410 455 10 j. j. PROPN bracis-28410 455 11 , , PUNCT bracis-28410 455 12 fan fan PROPN bracis-28410 455 13 , , PUNCT bracis-28410 455 14 y. y. PROPN bracis-28410 455 15 , , PUNCT bracis-28410 455 16 yang yang PROPN bracis-28410 455 17 , , PUNCT bracis-28410 455 18 j. j. PROPN bracis-28410 455 19 , , PUNCT bracis-28410 455 20 xu xu PROPN bracis-28410 455 21 , , PUNCT bracis-28410 455 22 n. n. PROPN bracis-28410 455 23 , , PUNCT bracis-28410 455 24 wang wang PROPN bracis-28410 455 25 , , PUNCT bracis-28410 455 26 x. x. PROPN bracis-28410 455 27 , , PUNCT bracis-28410 455 28 huang huang PROPN bracis-28410 455 29 , , PUNCT bracis-28410 455 30 t.s t.s PROPN bracis-28410 455 31 . . PROPN bracis-28410 455 32 : : PUNCT bracis-28410 455 33 wide wide ADJ bracis-28410 455 34 activation activation NOUN bracis-28410 455 35 for for ADP bracis-28410 455 36 efficient efficient ADJ bracis-28410 455 37 and and CCONJ bracis-28410 455 38 accurate accurate ADJ bracis-28410 455 39 image image NOUN bracis-28410 455 40 super super NOUN bracis-28410 455 41 - - NOUN bracis-28410 455 42 resolution resolution NOUN bracis-28410 455 43 . . PUNCT bracis-28410 456 1 arxiv arxiv PROPN bracis-28410 456 2 preprint preprint PROPN bracis-28410 456 3 arxiv:1808.08718 arxiv:1808.08718 NUM bracis-28410 456 4 ( ( PUNCT bracis-28410 456 5 2018 2018 NUM bracis-28410 456 6 ) ) PUNCT bracis-28410 456 7 zhang zhang PROPN bracis-28410 456 8 , , PUNCT bracis-28410 456 9 k. k. PROPN bracis-28410 456 10 , , PUNCT bracis-28410 456 11 gu gu PROPN bracis-28410 456 12 , , PUNCT bracis-28410 456 13 s. s. PROPN bracis-28410 456 14 , , PUNCT bracis-28410 456 15 timofte timofte PROPN bracis-28410 456 16 , , PUNCT bracis-28410 456 17 r. r. PROPN bracis-28410 456 18 : : PUNCT bracis-28410 456 19 ntire ntire NOUN bracis-28410 456 20 2020 2020 NUM bracis-28410 456 21 challenge challenge NOUN bracis-28410 456 22 on on ADP bracis-28410 456 23 perceptual perceptual ADJ bracis-28410 456 24 extreme extreme ADJ bracis-28410 456 25 super super NOUN bracis-28410 456 26 - - NOUN bracis-28410 456 27 resolution resolution NOUN bracis-28410 456 28 : : PUNCT bracis-28410 456 29 methods method NOUN bracis-28410 456 30 and and CCONJ bracis-28410 456 31 results result NOUN bracis-28410 456 32 . . PUNCT bracis-28410 457 1 in in ADP bracis-28410 457 2 : : PUNCT bracis-28410 457 3 cvpr cvpr NOUN bracis-28410 457 4 workshops workshop NOUN bracis-28410 457 5 , , PUNCT bracis-28410 457 6 pp pp ADP bracis-28410 457 7 . . PUNCT bracis-28410 458 1 1–10 1–10 NOUN bracis-28410 458 2 ( ( PUNCT bracis-28410 458 3 2020 2020 NUM bracis-28410 458 4 ) ) PUNCT bracis-28410 458 5 google google PROPN bracis-28410 458 6 scholar scholar NOUN bracis-28410 458 7     SPACE bracis-28410 458 8 zhang zhang PROPN bracis-28410 458 9 , , PUNCT bracis-28410 458 10 y. y. PROPN bracis-28410 458 11 , , PUNCT bracis-28410 458 12 li li PROPN bracis-28410 458 13 , , PUNCT bracis-28410 458 14 k. k. PROPN bracis-28410 458 15 , , PUNCT bracis-28410 458 16 li li PROPN bracis-28410 458 17 , , PUNCT bracis-28410 458 18 k. k. PROPN bracis-28410 458 19 , , PUNCT bracis-28410 458 20 wang wang PROPN bracis-28410 458 21 , , PUNCT bracis-28410 458 22 l. l. PROPN bracis-28410 458 23 , , PUNCT bracis-28410 458 24 zhong zhong PROPN bracis-28410 458 25 , , PUNCT bracis-28410 458 26 b. b. PROPN bracis-28410 458 27 , , PUNCT bracis-28410 458 28 fu fu PROPN bracis-28410 458 29 , , PUNCT bracis-28410 458 30 y. y. NOUN bracis-28410 458 31 : : PUNCT bracis-28410 458 32 image image NOUN bracis-28410 458 33 super super NOUN bracis-28410 458 34 - - NOUN bracis-28410 458 35 resolution resolution NOUN bracis-28410 458 36 using use VERB bracis-28410 458 37 very very ADV bracis-28410 458 38 deep deep ADJ bracis-28410 458 39 residual residual ADJ bracis-28410 458 40 channel channel NOUN bracis-28410 458 41 attention attention NOUN bracis-28410 458 42 networks network NOUN bracis-28410 458 43 . . PUNCT bracis-28410 459 1 in in ADP bracis-28410 459 2 : : PUNCT bracis-28410 459 3 eccv eccv ADJ bracis-28410 459 4 , , PUNCT bracis-28410 459 5 pp pp ADV bracis-28410 459 6 . . PUNCT bracis-28410 460 1 1–8 1–8 NUM bracis-28410 460 2 ( ( PUNCT bracis-28410 460 3 2018 2018 NUM bracis-28410 460 4 ) ) PUNCT bracis-28410 460 5 google google PROPN bracis-28410 460 6 scholar scholar NOUN bracis-28410 460 7     SPACE bracis-28410 460 8 zhang zhang PROPN bracis-28410 460 9 , , PUNCT bracis-28410 460 10 y. y. PROPN bracis-28410 460 11 , , PUNCT bracis-28410 460 12 tian tian PROPN bracis-28410 460 13 , , PUNCT bracis-28410 460 14 y. y. PROPN bracis-28410 460 15 , , PUNCT bracis-28410 460 16 kong kong PROPN bracis-28410 460 17 , , PUNCT bracis-28410 460 18 y. y. PROPN bracis-28410 460 19 , , PUNCT bracis-28410 460 20 zhong zhong PROPN bracis-28410 460 21 , , PUNCT bracis-28410 460 22 b. b. PROPN bracis-28410 460 23 , , PUNCT bracis-28410 460 24 fu fu PROPN bracis-28410 460 25 , , PUNCT bracis-28410 460 26 y. y. NOUN bracis-28410 460 27 : : PUNCT bracis-28410 460 28 residual residual ADJ bracis-28410 460 29 dense dense ADJ bracis-28410 460 30 network network NOUN bracis-28410 460 31 for for ADP bracis-28410 460 32 image image NOUN bracis-28410 460 33 super super NOUN bracis-28410 460 34 - - NOUN bracis-28410 460 35 resolution resolution NOUN bracis-28410 460 36 . . PUNCT bracis-28410 461 1 in in ADP bracis-28410 461 2 : : PUNCT bracis-28410 461 3 cvpr cvpr NOUN bracis-28410 461 4 , , PUNCT bracis-28410 461 5 pp pp ADV bracis-28410 461 6 . . PUNCT bracis-28410 462 1 2472–2481 2472–2481 NUM bracis-28410 462 2 ( ( PUNCT bracis-28410 462 3 2018 2018 NUM bracis-28410 462 4 ) ) PUNCT bracis-28410 462 5 google google PROPN bracis-28410 462 6 scholar scholar NOUN bracis-28410 462 7     SPACE bracis-28410 462 8 zhao zhao PROPN bracis-28410 462 9 , , PUNCT bracis-28410 462 10 h. h. PROPN bracis-28410 462 11 , , PUNCT bracis-28410 462 12 gallo gallo PROPN bracis-28410 462 13 , , PUNCT bracis-28410 462 14 o. o. PROPN bracis-28410 462 15 , , PUNCT bracis-28410 462 16 frosio frosio PROPN bracis-28410 462 17 , , PUNCT bracis-28410 462 18 i. i. PROPN bracis-28410 462 19 , , PUNCT bracis-28410 462 20 kautz kautz PROPN bracis-28410 462 21 , , PUNCT bracis-28410 462 22 j. j. PROPN bracis-28410 462 23 : : PUNCT bracis-28410 462 24 loss loss NOUN bracis-28410 462 25 functions function NOUN bracis-28410 462 26 for for ADP bracis-28410 462 27 image image NOUN bracis-28410 462 28 restoration restoration NOUN bracis-28410 462 29 with with ADP bracis-28410 462 30 neural neural ADJ bracis-28410 462 31 networks network NOUN bracis-28410 462 32 . . PUNCT bracis-28410 463 1 ieee ieee PROPN bracis-28410 463 2 trans trans PROPN bracis-28410 463 3 . . PUNCT bracis-28410 464 1 comput comput NOUN bracis-28410 464 2 . . PUNCT bracis-28410 465 1 imaging imaging NOUN bracis-28410 465 2 3(1 3(1 NUM bracis-28410 465 3 ) ) PUNCT bracis-28410 465 4 , , PUNCT bracis-28410 465 5 47–57 47–57 NUM bracis-28410 465 6 ( ( PUNCT bracis-28410 465 7 2017 2017 NUM bracis-28410 465 8 ) ) PUNCT bracis-28410 465 9 article article NOUN bracis-28410 465 10     SPACE bracis-28410 465 11 google google PROPN bracis-28410 465 12 scholar scholar NOUN bracis-28410 465 13     SPACE bracis-28410 465 14 download download NOUN bracis-28410 465 15 references reference NOUN bracis-28410 465 16 author author NOUN bracis-28410 465 17 information information NOUN bracis-28410 465 18 authors author NOUN bracis-28410 465 19 and and CCONJ bracis-28410 465 20 affiliations affiliation NOUN bracis-28410 465 21 institute institute PROPN bracis-28410 465 22 of of ADP bracis-28410 465 23 computing computing PROPN bracis-28410 465 24 , , PUNCT bracis-28410 465 25 university university PROPN bracis-28410 465 26 of of ADP bracis-28410 465 27 campinas campinas PROPN bracis-28410 465 28 , , PUNCT bracis-28410 465 29 campinas campinas PROPN bracis-28410 465 30 , , PUNCT bracis-28410 465 31 sp sp NOUN bracis-28410 465 32 , , PUNCT bracis-28410 465 33 13083 13083 NUM bracis-28410 465 34 - - SYM bracis-28410 465 35 852 852 NUM bracis-28410 465 36 , , PUNCT bracis-28410 465 37 brazil brazil PROPN bracis-28410 465 38 george george PROPN bracis-28410 465 39 corrêa corrêa PROPN bracis-28410 465 40 de de PROPN bracis-28410 465 41 araújo araújo PROPN bracis-28410 465 42     SPACE bracis-28410 465 43 & & CCONJ bracis-28410 465 44     SPACE bracis-28410 465 45 helio helio PROPN bracis-28410 465 46 pedrini pedrini PROPN bracis-28410 465 47 polytechnic polytechnic ADJ bracis-28410 465 48 school school NOUN bracis-28410 465 49 of of ADP bracis-28410 465 50 the the DET bracis-28410 465 51 university university PROPN bracis-28410 465 52 of of ADP bracis-28410 465 53 são são PROPN bracis-28410 465 54 paulo paulo PROPN bracis-28410 465 55 , , PUNCT bracis-28410 465 56 university university PROPN bracis-28410 465 57 of of ADP bracis-28410 465 58 são são PROPN bracis-28410 465 59 paulo paulo PROPN bracis-28410 465 60 , , PUNCT bracis-28410 465 61 são são PROPN bracis-28410 465 62 paulo paulo PROPN bracis-28410 465 63 , , PUNCT bracis-28410 465 64 sp sp NOUN bracis-28410 465 65 , , PUNCT bracis-28410 465 66 05508 05508 NUM bracis-28410 465 67 - - SYM bracis-28410 465 68 010 010 NUM bracis-28410 465 69 , , PUNCT bracis-28410 465 70 brazil brazil PROPN bracis-28410 465 71 artur artur PROPN bracis-28410 465 72 jordão jordão PROPN bracis-28410 465 73 authors author NOUN bracis-28410 465 74 george george PROPN bracis-28410 465 75 corrêa corrêa PROPN bracis-28410 465 76 de de PROPN bracis-28410 465 77 araújoview araújoview PROPN bracis-28410 465 78 author author NOUN bracis-28410 465 79 publications publication NOUN bracis-28410 465 80 search search NOUN bracis-28410 465 81 author author NOUN bracis-28410 465 82 on on ADP bracis-28410 465 83 : : PUNCT bracis-28410 465 84 pubmed pubmed PROPN bracis-28410 465 85     SPACE bracis-28410 465 86 google google PROPN bracis-28410 465 87 scholar scholar PROPN bracis-28410 465 88 artur artur PROPN bracis-28410 465 89 jordãoview jordãoview PROPN bracis-28410 465 90 author author NOUN bracis-28410 465 91 publications publication NOUN bracis-28410 465 92 search search NOUN bracis-28410 465 93 author author NOUN bracis-28410 465 94 on on ADP bracis-28410 465 95 : : PUNCT bracis-28410 465 96 pubmed pubmed PROPN bracis-28410 465 97     SPACE bracis-28410 465 98 google google PROPN bracis-28410 465 99 scholar scholar NOUN bracis-28410 465 100 helio helio NOUN bracis-28410 465 101 pedriniview pedriniview NOUN bracis-28410 465 102 author author NOUN bracis-28410 465 103 publications publication NOUN bracis-28410 465 104 search search NOUN bracis-28410 465 105 author author NOUN bracis-28410 465 106 on on ADP bracis-28410 465 107 : : PUNCT bracis-28410 465 108 pubmed pubmed PROPN bracis-28410 465 109     SPACE bracis-28410 465 110 google google PROPN bracis-28410 465 111 scholar scholar NOUN bracis-28410 465 112 corresponding correspond VERB bracis-28410 465 113 author author NOUN bracis-28410 465 114 correspondence correspondence NOUN bracis-28410 465 115 to to ADP bracis-28410 465 116 artur artur PROPN bracis-28410 465 117 jordão jordão PROPN bracis-28410 465 118 . . PUNCT bracis-28410 466 1 editor editor NOUN bracis-28410 466 2 information information NOUN bracis-28410 466 3 editors editor NOUN bracis-28410 466 4 and and CCONJ bracis-28410 466 5 affiliations affiliation NOUN bracis-28410 466 6 federal federal PROPN bracis-28410 466 7 university university PROPN bracis-28410 466 8 of of ADP bracis-28410 466 9 são são PROPN bracis-28410 466 10 carlos carlos PROPN bracis-28410 466 11 , , PUNCT bracis-28410 466 12 são são PROPN bracis-28410 466 13 carlos carlos PROPN bracis-28410 466 14 , , PUNCT bracis-28410 466 15 brazil brazil PROPN bracis-28410 466 16 murilo murilo PROPN bracis-28410 466 17 c. c. PROPN bracis-28410 466 18 naldi naldi PROPN bracis-28410 466 19 centro centro PROPN bracis-28410 466 20 universitario universitario PROPN bracis-28410 466 21 da da PROPN bracis-28410 466 22 fei fei PROPN bracis-28410 466 23 , , PUNCT bracis-28410 466 24 são são PROPN bracis-28410 466 25 bernardo bernardo PROPN bracis-28410 466 26 do do AUX bracis-28410 466 27 campo campo PROPN bracis-28410 466 28 , , PUNCT bracis-28410 466 29 brazil brazil PROPN bracis-28410 466 30 reinaldo reinaldo PROPN bracis-28410 466 31 a. a. PROPN bracis-28410 466 32 c. c. PROPN bracis-28410 466 33 bianchi bianchi PROPN bracis-28410 466 34 rights right NOUN bracis-28410 466 35 and and CCONJ bracis-28410 466 36 permissions permission NOUN bracis-28410 466 37 reprints reprint NOUN bracis-28410 466 38 and and CCONJ bracis-28410 466 39 permissions permission VERB bracis-28410 466 40 copyright copyright NOUN bracis-28410 466 41 information information NOUN bracis-28410 466 42 © © ADP bracis-28410 466 43 2023 2023 NUM bracis-28410 466 44 the the DET bracis-28410 466 45 author(s author(s NOUN bracis-28410 466 46 ) ) PUNCT bracis-28410 466 47 , , PUNCT bracis-28410 466 48 under under ADP bracis-28410 466 49 exclusive exclusive ADJ bracis-28410 466 50 license license NOUN bracis-28410 466 51 to to ADP bracis-28410 466 52 springer springer NOUN bracis-28410 466 53 nature nature PROPN bracis-28410 466 54 switzerland switzerland PROPN bracis-28410 466 55 ag ag PROPN bracis-28410 466 56 about about ADP bracis-28410 466 57 this this DET bracis-28410 466 58 paper paper NOUN bracis-28410 466 59 cite cite VERB bracis-28410 466 60 this this DET bracis-28410 466 61 paper paper NOUN bracis-28410 466 62 de de ADP bracis-28410 466 63 araújo araújo PROPN bracis-28410 466 64 , , PUNCT bracis-28410 466 65 g.c g.c PROPN bracis-28410 466 66 . . PROPN bracis-28410 466 67 , , PUNCT bracis-28410 466 68 jordão jordão PROPN bracis-28410 466 69 , , PUNCT bracis-28410 466 70 a. a. NOUN bracis-28410 466 71 , , PUNCT bracis-28410 466 72 pedrini pedrini PROPN bracis-28410 466 73 , , PUNCT bracis-28410 466 74 h. h. PROPN bracis-28410 466 75 ( ( PUNCT bracis-28410 466 76 2023 2023 NUM bracis-28410 466 77 ) ) PUNCT bracis-28410 466 78 . . PUNCT bracis-28410 467 1 single single ADJ bracis-28410 467 2 image image NOUN bracis-28410 467 3 super super NOUN bracis-28410 467 4 - - NOUN bracis-28410 467 5 resolution resolution NOUN bracis-28410 467 6 based base VERB bracis-28410 467 7 on on ADP bracis-28410 467 8     SPACE bracis-28410 467 9 capsule capsule NOUN bracis-28410 467 10 neural neural ADJ bracis-28410 467 11 networks network NOUN bracis-28410 467 12 . . PUNCT bracis-28410 468 1 in in ADP bracis-28410 468 2 : : PUNCT bracis-28410 468 3 naldi naldi PROPN bracis-28410 468 4 , , PUNCT bracis-28410 468 5 m.c m.c PROPN bracis-28410 468 6 . . PROPN bracis-28410 468 7 , , PUNCT bracis-28410 468 8 bianchi bianchi PROPN bracis-28410 468 9 , , PUNCT bracis-28410 468 10 r.a.c r.a.c ADP bracis-28410 468 11 . . PUNCT bracis-28410 468 12 ( ( PUNCT bracis-28410 468 13 eds ed NOUN bracis-28410 468 14 ) ) PUNCT bracis-28410 468 15 intelligent intelligent ADJ bracis-28410 468 16 systems system NOUN bracis-28410 468 17 . . PUNCT bracis-28410 469 1 bracis bracis PROPN bracis-28410 469 2 2023 2023 NUM bracis-28410 469 3 . . PUNCT bracis-28410 470 1 lecture lecture NOUN bracis-28410 470 2 notes note NOUN bracis-28410 470 3 in in ADP bracis-28410 470 4 computer computer NOUN bracis-28410 470 5 science science NOUN bracis-28410 470 6 ( ( PUNCT bracis-28410 470 7 ) ) PUNCT bracis-28410 470 8 , , PUNCT bracis-28410 470 9 vol vol NOUN bracis-28410 470 10 14197 14197 NUM bracis-28410 470 11 . . PUNCT bracis-28410 471 1 springer springer NOUN bracis-28410 471 2 , , PUNCT bracis-28410 471 3 cham cham PROPN bracis-28410 471 4 . . PUNCT bracis-28410 472 1 https://doi.org/10.1007/978-3-031-45392-2_8 https://doi.org/10.1007/978-3-031-45392-2_8 PROPN bracis-28410 472 2 download download PROPN bracis-28410 472 3 citation citation NOUN bracis-28410 472 4 .ris .ris PUNCT bracis-28410 473 1 .enw .enw PROPN bracis-28410 473 2 .bib .bib PUNCT bracis-28410 474 1 doi doi PROPN bracis-28410 474 2 : : PUNCT bracis-28410 474 3 https://doi.org/10.1007/978-3-031-45392-2_8 https://doi.org/10.1007/978-3-031-45392-2_8 NOUN bracis-28410 474 4 published publish VERB bracis-28410 474 5 : : PUNCT bracis-28410 474 6 12 12 NUM bracis-28410 474 7 october october PROPN bracis-28410 474 8 2023 2023 NUM bracis-28410 474 9 publisher publisher NOUN bracis-28410 474 10 name name NOUN bracis-28410 474 11 : : PUNCT bracis-28410 474 12 springer springer NOUN bracis-28410 474 13 , , PUNCT bracis-28410 474 14 cham cham PROPN bracis-28410 474 15 print print PROPN bracis-28410 474 16 isbn isbn PROPN bracis-28410 474 17 : : PUNCT bracis-28410 474 18 978 978 NUM bracis-28410 474 19 - - SYM bracis-28410 474 20 3 3 NUM bracis-28410 474 21 - - PUNCT bracis-28410 474 22 031 031 NUM bracis-28410 474 23 - - PUNCT bracis-28410 474 24 45391 45391 NUM bracis-28410 474 25 - - SYM bracis-28410 474 26 5 5 NUM bracis-28410 474 27 online online ADJ bracis-28410 474 28 isbn isbn NOUN bracis-28410 474 29 : : PUNCT bracis-28410 474 30 978 978 NUM bracis-28410 474 31 - - SYM bracis-28410 474 32 3 3 NUM bracis-28410 474 33 - - PUNCT bracis-28410 474 34 031 031 NUM bracis-28410 474 35 - - PUNCT bracis-28410 474 36 45392 45392 NUM bracis-28410 474 37 - - SYM bracis-28410 474 38 2 2 NUM bracis-28410 474 39 ebook ebook NOUN bracis-28410 474 40 packages package NOUN bracis-28410 474 41 : : PUNCT bracis-28410 474 42 computer computer NOUN bracis-28410 474 43 sciencecomputer sciencecomputer NOUN bracis-28410 474 44 science science NOUN bracis-28410 474 45 ( ( PUNCT bracis-28410 474 46 r0 r0 NOUN bracis-28410 474 47 ) ) PUNCT bracis-28410 474 48 share share VERB bracis-28410 474 49 this this DET bracis-28410 474 50 paper paper NOUN bracis-28410 474 51 anyone anyone PRON bracis-28410 474 52 you you PRON bracis-28410 474 53 share share VERB bracis-28410 474 54 the the DET bracis-28410 474 55 following follow VERB bracis-28410 474 56 link link NOUN bracis-28410 474 57 with with ADP bracis-28410 474 58 will will AUX bracis-28410 474 59 be be AUX bracis-28410 474 60 able able ADJ bracis-28410 474 61 to to PART bracis-28410 474 62 read read VERB bracis-28410 474 63 this this DET bracis-28410 474 64 content content NOUN bracis-28410 474 65 : : PUNCT bracis-28410 474 66 get get VERB bracis-28410 474 67 shareable shareable ADJ bracis-28410 474 68 linksorry linksorry NOUN bracis-28410 474 69 , , PUNCT bracis-28410 474 70 a a DET bracis-28410 474 71 shareable shareable ADJ bracis-28410 474 72 link link NOUN bracis-28410 474 73 is be AUX bracis-28410 474 74 not not PART bracis-28410 474 75 currently currently ADV bracis-28410 474 76 available available ADJ bracis-28410 474 77 for for ADP bracis-28410 474 78 this this DET bracis-28410 474 79 article article NOUN bracis-28410 474 80 . . PUNCT bracis-28410 475 1 copy copy VERB bracis-28410 475 2 shareable shareable ADJ bracis-28410 475 3 link link NOUN bracis-28410 475 4 to to PART bracis-28410 475 5 clipboard clipboard NOUN bracis-28410 475 6 provided provide VERB bracis-28410 475 7 by by ADP bracis-28410 475 8 the the DET bracis-28410 475 9 springer springer NOUN bracis-28410 475 10 nature nature PROPN bracis-28410 475 11 sharedit sharedit PROPN bracis-28410 475 12 content content NOUN bracis-28410 475 13 - - PUNCT bracis-28410 475 14 sharing share VERB bracis-28410 475 15 initiative initiative NOUN bracis-28410 475 16 keywords keyword VERB bracis-28410 475 17 neural neural ADJ bracis-28410 475 18 network network NOUN bracis-28410 475 19 capsule capsule PROPN bracis-28410 475 20 networks network NOUN bracis-28410 475 21 image image VERB bracis-28410 475 22 super super ADJ bracis-28410 475 23 - - ADJ bracis-28410 475 24 resolution resolution NOUN bracis-28410 475 25 publish publish NOUN bracis-28410 475 26 with with ADP bracis-28410 475 27 us us PROPN bracis-28410 475 28 policies policy NOUN bracis-28410 475 29 and and CCONJ bracis-28410 475 30 ethics ethic NOUN bracis-28410 475 31 search search NOUN bracis-28410 475 32 search search NOUN bracis-28410 475 33 by by ADP bracis-28410 475 34 keyword keyword NOUN bracis-28410 475 35 or or CCONJ bracis-28410 475 36 author author NOUN bracis-28410 475 37 search search NOUN bracis-28410 475 38 navigation navigation NOUN bracis-28410 475 39 find find VERB bracis-28410 475 40 a a DET bracis-28410 475 41 journal journal NOUN bracis-28410 475 42 publish publish VERB bracis-28410 475 43 with with ADP bracis-28410 475 44 us we PRON bracis-28410 475 45 track track VERB bracis-28410 475 46 your your PRON bracis-28410 475 47 research research NOUN bracis-28410 475 48 discover discover VERB bracis-28410 475 49 content content NOUN bracis-28410 475 50 journals journal NOUN bracis-28410 475 51 a a DET bracis-28410 475 52 - - PUNCT bracis-28410 475 53 z z NOUN bracis-28410 475 54 books book NOUN bracis-28410 475 55 a a DET bracis-28410 475 56 - - PUNCT bracis-28410 475 57 z z NOUN bracis-28410 475 58 publish publish NOUN bracis-28410 475 59 with with ADP bracis-28410 475 60 us us PROPN bracis-28410 475 61 journal journal PROPN bracis-28410 475 62 finder finder PROPN bracis-28410 475 63 publish publish VERB bracis-28410 475 64 your your PRON bracis-28410 475 65 research research NOUN bracis-28410 475 66 language language NOUN bracis-28410 475 67 editing edit VERB bracis-28410 475 68 open open ADJ bracis-28410 475 69 access access NOUN bracis-28410 475 70 publishing publishing NOUN bracis-28410 475 71 products product NOUN bracis-28410 475 72 and and CCONJ bracis-28410 475 73 services service NOUN bracis-28410 475 74 our our PRON bracis-28410 475 75 products product NOUN bracis-28410 475 76 librarians librarian VERB bracis-28410 475 77 societies society NOUN bracis-28410 475 78 partners partner NOUN bracis-28410 475 79 and and CCONJ bracis-28410 475 80 advertisers advertiser NOUN bracis-28410 475 81 our our PRON bracis-28410 475 82 brands brand NOUN bracis-28410 475 83 springer springer NOUN bracis-28410 475 84 nature nature PROPN bracis-28410 475 85 portfolio portfolio PROPN bracis-28410 475 86 bmc bmc PROPN bracis-28410 475 87 palgrave palgrave PROPN bracis-28410 475 88 macmillan macmillan PROPN bracis-28410 475 89 apress apress PROPN bracis-28410 475 90 discover discover VERB bracis-28410 475 91 your your PRON bracis-28410 475 92 privacy privacy NOUN bracis-28410 475 93 choices choice NOUN bracis-28410 475 94 / / SYM bracis-28410 475 95 manage manage NOUN bracis-28410 475 96 cookies cookie NOUN bracis-28410 475 97 your your PRON bracis-28410 475 98 us us PROPN bracis-28410 476 1 state state NOUN bracis-28410 476 2 privacy privacy NOUN bracis-28410 476 3 rights right NOUN bracis-28410 476 4 accessibility accessibility NOUN bracis-28410 476 5 statement statement NOUN bracis-28410 476 6 terms term NOUN bracis-28410 476 7 and and CCONJ bracis-28410 476 8 conditions condition NOUN bracis-28410 476 9 privacy privacy NOUN bracis-28410 476 10 policy policy NOUN bracis-28410 476 11 help help NOUN bracis-28410 476 12 and and CCONJ bracis-28410 476 13 support support VERB bracis-28410 476 14 legal legal ADJ bracis-28410 476 15 notice notice NOUN bracis-28410 476 16 cancel cancel VERB bracis-28410 476 17 contracts contract NOUN bracis-28410 476 18 here here ADV bracis-28410 476 19 129.74.145.123 129.74.145.123 NUM bracis-28410 476 20 hesburgh hesburgh PROPN bracis-28410 476 21 library library PROPN bracis-28410 476 22 er er INTJ bracis-28410 476 23 unit unit NOUN bracis-28410 476 24 ( ( PUNCT bracis-28410 476 25 3005732405 3005732405 NUM bracis-28410 476 26 ) ) PUNCT bracis-28410 476 27 northeast northeast ADJ bracis-28410 476 28 research research NOUN bracis-28410 476 29 libraries library NOUN bracis-28410 476 30 ( ( PUNCT bracis-28410 476 31 nerl nerl PROPN bracis-28410 476 32 ) ) PUNCT bracis-28410 476 33 ( ( PUNCT bracis-28410 476 34 8200828607 8200828607 NUM bracis-28410 476 35 ) ) PUNCT bracis-28410 476 36 nerl nerl VERB bracis-28410 476 37 ta ta X bracis-28410 476 38 account account NOUN bracis-28410 476 39 ( ( PUNCT bracis-28410 476 40 3006206169 3006206169 NUM bracis-28410 476 41 ) ) PUNCT bracis-28410 476 42 university university NOUN bracis-28410 476 43 of of ADP bracis-28410 476 44 notre notre PROPN bracis-28410 476 45 dame dame PROPN bracis-28410 476 46 hesburgh hesburgh PROPN bracis-28410 476 47 library library NOUN bracis-28410 476 48 ( ( PUNCT bracis-28410 476 49 3000184373 3000184373 NUM bracis-28410 476 50 ) ) PUNCT bracis-28410 477 1 © © ADP bracis-28410 477 2 2025 2025 NUM bracis-28410 477 3 springer springer NOUN bracis-28410 477 4 nature nature NOUN